View
0
Download
0
Category
Preview:
Citation preview
An Efficiency Study of the
Australian Superannuation
Industry
Jason Zhiwei Qu
16 February 2014
Disclaimer and Copyright
The material in this report is copyright of Jason Zhiwei Qu. The views and opinions expressed in this report are solely that of the author’s and do not reflect the views and opinions of the Australian Prudential Regulation Authority. Any errors in this report are the responsibility of the author. The material in this report is copyright. Other than for any use permitted under the Copyright Act 1968, all other rights are reserved and permission should be sought through the author prior to any reproduction.
Brian Gray Scholarship
Final Report
An Efficiency Study of
The Australian Superannuation Industry
Author:
Jason Zhiwei Qu
Student ID:
3334506
Supervisor:
Assoc.Prof. Hazel Bateman
Dr. Sasan Bakhtiari
Bachelor of Commerce / Bachelor of Economics
16th February, 2014
Declaration
I declare that this thesis is my own work and that, to the best of my knowledge, it
contains no material that has been published or written by another person(s) except
where due acknowledgment has been made. This thesis has not been submitted for
award of any other degree or diploma at the University of New South Wales or at any
other educational institution. I declare that the intellectual content of this thesis
is the product of my own work except to the extent that assistance from others is
acknowledged.
...................................
Jason Zhiwei Qu
16th February, 2014
i
Contents
Table of Contents i
Abstract 1
1 Introduction 2
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2 Motivation and Contributions . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Research Questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Institutional Setting 7
2.1 A Brief History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Current Landscape . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.1 Industry Composition . . . . . . . . . . . . . . . . . . . . . . 9
2.2.2 Coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.3 Benefit Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.4 Regulatory Framework . . . . . . . . . . . . . . . . . . . . . . 14
2.3 Industry Trends . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.3.1 Decreasing Levels of Competition . . . . . . . . . . . . . . . . 15
2.3.2 Stagnant Expenses . . . . . . . . . . . . . . . . . . . . . . . . 17
3 Literature Review 19
3.1 The Cooper Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.1.1 Reducing Costs . . . . . . . . . . . . . . . . . . . . . . . . . . 21
ii
3.1.2 Lower profit margins . . . . . . . . . . . . . . . . . . . . . . . 21
3.1.3 Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.1.4 Default Portfolios . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.1.5 Regulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2 International Literature . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2.1 Efficiency Studies: General Applications . . . . . . . . . . . . 24
3.2.2 Efficiency Studies: Pension Funds . . . . . . . . . . . . . . . . 28
3.3 Australian Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.4 Investment Performances and Risk . . . . . . . . . . . . . . . . . . . 31
3.5 Administrative Expenses . . . . . . . . . . . . . . . . . . . . . . . . . 34
4 Theoretical Framework 37
4.1 Efficiency Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.1.1 Farrell Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.1.2 Scale Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.2 Efficiency Score Estimation . . . . . . . . . . . . . . . . . . . . . . . 43
4.2.1 Model Selection . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.2.2 Limitations of DEA . . . . . . . . . . . . . . . . . . . . . . . . 45
4.2.3 Data Envelopment Analysis . . . . . . . . . . . . . . . . . . . 47
5 Empirical Specifications 52
5.1 Output Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.1.1 Administration . . . . . . . . . . . . . . . . . . . . . . . . . . 53
5.1.2 Investment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.1.3 Risk Management . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.2 Input Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.2.1 Total Assets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
5.2.2 Administrative Expenses . . . . . . . . . . . . . . . . . . . . . 59
6 Data 61
6.1 Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
iii
6.1.1 Financials Information . . . . . . . . . . . . . . . . . . . . . . 61
6.1.2 Structural Characteristics . . . . . . . . . . . . . . . . . . . . 63
6.2 Sample Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.2.1 Erroneous Data . . . . . . . . . . . . . . . . . . . . . . . . . . 65
6.3 Final data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
7 Results 67
7.1 Limitation of Inference . . . . . . . . . . . . . . . . . . . . . . . . . . 67
7.2 First Stage: Efficiency Scores Analysis . . . . . . . . . . . . . . . . . 68
7.2.1 Efficiency Rankings . . . . . . . . . . . . . . . . . . . . . . . . 71
7.2.2 Public Offering . . . . . . . . . . . . . . . . . . . . . . . . . . 76
7.2.3 Fund Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
7.2.4 Efficient Market . . . . . . . . . . . . . . . . . . . . . . . . . . 79
7.3 Second Stage: Regression Analysis . . . . . . . . . . . . . . . . . . . 82
7.3.1 Regressors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
7.3.2 Model Selection . . . . . . . . . . . . . . . . . . . . . . . . . . 85
7.3.3 Discussion of Results . . . . . . . . . . . . . . . . . . . . . . . 98
7.4 Policy Implication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.4.1 Extension: Aggregate Performance . . . . . . . . . . . . . . . 106
7.4.2 Key Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
8 Conclusion 112
iv
List of Figures
2.1 Assets Under Management: Fund Type . . . . . . . . . . . . . . . . . 11
2.2 Market Concentration . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Sector Decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.4 Average Fund Asset and Expense Ratio . . . . . . . . . . . . . . . . . 17
2.5 Average Expense Ratio by Fund Type . . . . . . . . . . . . . . . . . 18
4.1 Input-oriented efficiency measurement and the isoquant . . . . . . . . 38
4.2 The effect of Scale Efficiency . . . . . . . . . . . . . . . . . . . . . . . 41
4.3 Scale Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.1 Value at Risk Comparisons . . . . . . . . . . . . . . . . . . . . . . . . 57
7.1 Average Pure Technical Efficiency: Fund Type . . . . . . . . . . . . . 78
7.2 Average Pure Technical Efficiency: Market Exits . . . . . . . . . . . . 80
7.3 Standard Fractional Regression Models . . . . . . . . . . . . . . . . . 93
v
List of Tables
2.1 Composition of the Australian Superannuation Industry 2012 . . . . . 10
2.2 Superannuation Coverage (1982 - 2012) . . . . . . . . . . . . . . . . . 12
2.3 Membership Distribution: Benefit Type . . . . . . . . . . . . . . . . . 13
3.1 Research Areas - Berger and Humphrey (1997) . . . . . . . . . . . . . 24
3.2 Negative Sharpe Ratio Example - Sy and Liu (2009) . . . . . . . . . . 33
6.1 Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
7.1 Average Efficiency Scores . . . . . . . . . . . . . . . . . . . . . . . . . 69
7.2 Reference Decision Making Units for Unisuper . . . . . . . . . . . . 73
7.3 Average Pure Technical Efficiency: Public Offering . . . . . . . . . . 77
7.4 Efficiency Regression: LPM & TOBIT . . . . . . . . . . . . . . . . . 90
7.5 FRM Results with Four Link Function Specifications . . . . . . . . . 94
7.6 RESET Test Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
7.7 Information Criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
7.8 FRM (loglog) Partial Effects . . . . . . . . . . . . . . . . . . . . . . . 97
7.9 Net APE of 1% change in market share . . . . . . . . . . . . . . . . . 101
7.10 Average Efficiency Scores: Aggregate Performance . . . . . . . . . . . 107
7.11 FRM (cloglog) Partial Effects: Aggregate Performance . . . . . . . . 108
8.1 Average Efficiency by Fund Type . . . . . . . . . . . . . . . . . . . . 121
8.2 Return Volatility Benchmark Index . . . . . . . . . . . . . . . . . . . 122
8.3 Benchmark Return Volatilities (Std.Dev) . . . . . . . . . . . . . . . . 122
vi
8.4 OLS & TOBIT estimates . . . . . . . . . . . . . . . . . . . . . . . . . 123
8.5 Generalised Linear Model Results . . . . . . . . . . . . . . . . . . . . 124
vii
Abstract
This research analyses the efficiency of Australian superannuation industry between
2004 and 2012. A review of the Australia’s superannuation industry and the
surrounding literature is undertaken, as is a discussion of the techniques of efficiency
measurement. In this respect, the first part of this study provides an introduction
into superannuation fund efficiency measurements.
The empirical analysis was conducted using a two-stage procedure. In the first stage
Data Envelopment Analysis (DEA) is used to construct an efficient frontier based on
annual fund level data from the Australia Prudential Regulatory Authority (APRA).
The second stage calibrates a non-linear Frational Response Model (FRM) under
the GLM regression framework.
Our analysis contributes to the broader literature in three ways. First, a
comprehensive frontier efficiency study incorporating all fund types is unique in the
context of Australia’s superannuation industry. Second, we incorporate age profiles
of members in our key performance indicator; this is an innovative technique that
has not been adopted in the broader pension fund efficiency literature. Lastly, we
calibrate more vigorous second stage techniques in estimating the relative impacts
of structural variables, which is somewhat lacking in the existing study of super fund
efficiency in Australia.
Finally, our results indicate that overall, the Australian superannuation industry is
relatively efficient. However, there are significant caveats in the structural features
of the industry which prevents the full benefit of economies of scale from being
utilised. These result have significant implications for future industry structural
reforms and competition regulation.
1
Chapter 1
Introduction
1.1 Background
‘Pension policy has become one of the more volatile areas of economic reform in
recent years. The onset of demographic transition, in developing and developed
economies alike, has combined with concerns about the efficiency effects of a large
public sector, to generate a search for pension reform options which reduce the legal
responsibility of governments to provide financial support for the retired. Many
countries around the world have recently undertaken reform, are in the middle of
the reform process, or are actively debating reform options.’ - Bateman et al. (2001)
The need for pension policy reform goes beyond fiscal sustainability and economic
stability, as the quality of life for the retied is partly determined by their level
of income and wealth. Australia’s current taxpayer funded and means tested age
pension system has been designed to target low income populations and support a
basic, acceptable standard of living, accounting for prevailing community standards.
The general desire for higher standards of retirement living along with the pursuit
of fiscal sustainability of public pension scheme has led to wide spread legislative
reforms in Australia’s retirement income provision policies. The current policy
emphasis on publicly mandated and privately managed superannuation savings is
the core legacy of this reform.
Since the inception of the superannuation guarantee (SG) in 1992, the superan-
nuation system has undergone significant changes. Today the industry manages
over $1.6 trillion in assets, more than Australia’s annual national GDP, and is
projected to rise to $6.1 trillion by 2035.1 A recent ABS survey2 shows 43%
1Treasury Forecast 20122ABS Australian Social Trends 4102.0 (2009)
2
of retired Australians have somewhat benefited from superannuation, of which
20% solely depend on their superannuation balance for retirement. This figure
is expected to increase exponentially over the next decade. The issue is further
exacerbated by the transition of investment risk, as the industry has moved away
from Defined Benefit (DB) to Defined Contribution (DC) schemes in recent years.
As of 2012, over 82.6% of superannuation assets held in large super funds3 have
been allocated to accumulation plans, representing retirement savings of more than
93.1% of members.4 The rapid transition from the reliance on public pensions to
privately managed savings coupled with the shift in investment risk from employers
to members has placed greater responsibility on the superannuation fund custodians
to provides best value for money investment management that promotes the financial
adequacy and stability of its members in retirement.
1.2 Motivation and Contributions
The primary motivation for this research derives from the 2012 Cooper Review
of the superannuation industry also known as the Super System Review. The
Cooper Review highlighted the importance of examining the operating efficiency
of the industry. In particular, whether the full benefit of economies of scale is
being harnessed and whether processes and procedures are still appropriate given
the changes that have occurred in the industry over the last decade. This research
aims to contribute to this debate by analysing the industry’s production efficiency
and the relative impact of key structural and design features. Given the diversity
of fund structures and plan features that operate concurrently in the Australian
market, the potential findings of systematic inefficiencies within the industry will
better inform future policy formulation.
Furthermore, previous research has shown that the level of administrative fees
prevalent in the APRA regulated superannuation funds5 have been very persistent
3Funds with more than four members4APRA Annual Superannuation Bulletin 20135Australian Prudential Regulatory Authority (APRA) is the main regulator in Australian
3
while returns have not improved, despite rapid consolidation of funds occurring
within the superannuation industry (Bateman, 2002). This is in contradiction with
conventional economic wisdom which suggests that increases in economies of scale
would likely result in the lowering of administrative fees due to greater dispersion of
fixed costs. Existing literature has been inadequate in addressing this contradiction,
with much of the studies focusing partially or entirely on investment performance
indicators such as return, risk and expenses. Our research seeks to provide some
insights to this research gap by examining the overall production efficiency of super
funds in delivering its core services to members.
A review of international pension fund efficiency literature revealed that these
studies are mostly conservative, which rarely departs from traditional banking and
mutual fund output/input specifications. This study argues that such practice
is inadequate in capturing the unique features of retirement income provision, in
particular, the fixed statutory holding period faced by members.6 Our research
seeks to incorporate the age profile of members in the key performance indicator of
super funds. This technique is innovative and more accurately depicts the distinction
between super funds and other investment services.
Previous study of super fund efficiency in Australia (Sathye, 2011) has adopted a
similar two-stage DEA approach. However, we argue that the study was inadequate
in addressing the prevailing concerns regarding the misspecification of conventional
linear models in the second stage. This research seeks to address this issue by
calibrating non-linear models that is increasingly prevalent in the broader DEA
literature.
superannuation industry, it’s regulatory jurisdiction include all super funds with more than fourmembers. The rest of the industry composed of exempted public sector funds and Self ManagedSuper Funds (SMSF) which are regulated by the Australian Taxation Office (ATO)
6In the Australian context, members are forbidden from accessing their retirement saving beforethe age of 60 by law.
4
1.3 Research Questions
This paper aims to answer the following questions:
• What is the appropriate technique to measure the efficiency of the Australian
superannuation industry? How should these models be calibrated to accurately
reflect the industry’s unique functionalities?
• What are the impacts of key structural and design features on the efficiency
of super funds?
• What are the implications of future retirement policy reform?
1.4 Structure
This research is organised in the following manner:
Chapter 2 provides an overview of the superannuation industry in Australia,
including its structural characteristics, design features and industry trends. This
chapter serves as a background to our current research.
Chapter 3 presents a summary of literature surrounding this research. This includes
the broader international literature on the efficiency of financial service sectors
and existing superannuation performance studies in Australia. The chapter will
seeks to identify key research gaps and provides some basis for our empirical model
calibration.
Chapter 4 briefly discusses the theoretical framework behind efficiency measure-
ments and the DEA technique. More importantly, the chapter outlines the key
consideration one must take in implementing an efficiency study.
Chapter 5 is concerned with the DEA input/output calibration. A brief discussion
of the underlying theory of variable selection is also provided.
5
Chapter 6 discusses aspects of data handling, including sample selection, the
construction of output/input measures and the potential contextual variables.
Chapter 7 present the results of this research. First, we investigate the implications
of first stage results, especially in terms of average efficiency and scale efficiency.
Next, the chapter provides an overview of popular second stage regression models
and calibrate these to our analysis. The implications of our second stage results are
also discussed.
Chapter 8 concludes the study, where key results and policy implications are
summarised.
6
Chapter 2
Institutional Setting
In contrast to the private pension industry in comparable countries, the Australian
pension environment permits considerable diversity of plan design, structure and
size. These unique features of Australian superannuation (pension) funds 1 are
inextricably linked the history of pension policy reforms in Australia. Section 2.1
briefly discusses the history of pension policy reforms in Australia. Section 2.2
describes the current institutional framework that governs superannuation funds in
Australia. Section 2.3 outlines some puzzling empirical observations that motivate
this research.
2.1 A Brief History
Private pension schemes first emerged in Australia in the mid 19th century. Since
then, their key functionality and coverage evolved markedly up to the time of the
introduction of mandatory superannuation in 1992. The development of pension
policy in Australia can be broadly divided into three periods. During the first
era, until the 1940s, occupational superannuation provided a selected group of
salaried employees with an independent source of retirement income. The second
era, from 1950s to 1970s saw the relaxation of the means testing of the age pension,
and superannuation acted as supplement the age pension for mostly white-collar
workers. From the 1970s until the introduction of award superannuation in 1986
and subsequently mandatory occupational superannuation in 1992, superannuation
was an employment fringe benefit which, although more generally available, was
still concentrated among professional, managers and administrators, public sector
employees, and workers in the financial sector (Kewley (1973)).
1Superannuation is the Australian term for private pensions and that the two term are usedinterchangeably
7
Policy reforms initiated by government inquiries and industry movements have
helped shaped the current structure of the Australian superannuation industry
today. Notably, the Royal Commission established by the Bruce government
in 1923 was the first to consider extending the limited coverage of the existing
contributory retirement benefit schemes. However, it was the minority findings of the
Hancock review commissioned by the Whitlam government in 1973 which sparked
a shift in the emphasis of retirement income policy away from poverty alleviation
through the minimalist age pension towards income maintenance via contributory
superannuation. The impetus for universal contributory superannuation came
primarily through the industrial relations arena in the 1970s and 1980s. For the
union movement, the existing occupational superannuation available to middle to
senior management provided a platforms upon which workers could obtain deferred
wage increases in the form of retirement savings without going outside the bounds
of Australia’s then centralised wage fixing system (Kewley (1973)).
Then, in the mid 1980s the Hawke labor government included superannuation in
its contract (or The Accord) with the union movement. This saw the introduction
of award superannuation arrangements that constituted the first working version
of a mandatory private saving framework in Australia. Ultimately, this led to the
introduction of the Superannuation Guarantee (SG) in 1992, which is the basis of
the current superannuation landscape in Australia.
In its first phase, the SG increased from 4% to 9% between 1992 and 2002. Under
more recent reforms, employers must now contribute at least 9.25% of payroll
on behalf of their employees to a funded pension program, increasing to 12% by
2019(Bateman and Mitchell (2001)).
8
2.2 Current Landscape
Current retirement income provision in Australia is a three pillar construct,
comprising a public pension (the Age Pension), mandatory private saving (the
Superannuation Guarantee) and voluntary long-term savings including voluntary
superannuation. Given the reality of the ageing population in Australia, the efficacy
of the second pillar has become increasingly important in ensuring the long term
viability of the means tested Age Pension due to increasing upward pressure on
future pension liabilities. This section will summarise the current characteristics of
the mandatory superannuation environment in Australia.
2.2.1 Industry Composition
The diversity of funds in the current Australian Superannuation Industry is a
legacy of its development process prior to the introduction of the Superannuation
Guarantee. Australian superannuation funds (super funds) are privately managed
trusts with different benefit designs, profit orientations and operational structures.
The industry can be broadly categorised into two regulatory jurisdictions: (1)
Institutional superannuation funds that are governed by trustees on behalf of
members and are under the jurisdiction of the Australian Prudential Regulatory
Authority (APRA). These account for 59.5% of industry assets; and (2) Self-
Managed Superannuation Funds (SMSF)2 are regulated by the Australian Taxation
Office (ATO). These account for 31.5% of industry assets; The rest of the industry
comprises of public sector funds and life office statutory funds which are exempt
from regulations, accounting for 9% of total superannuation assets in Australia.
Institutional superannuation funds can be further categorised by their functional
2SMSF are funds with less than five members, all of which are trustees of the fund. Reportingobligations of SMSF are to the ATO, these informations are not publicly available.
9
classification. That is, (1) Corporate funds ; (2) Industry funds; (3) Public Sector
funds; (4) Retail funds. Superannuation funds in categories (1)-(3) are typically
’not for profit’ funds that cater only to members in their respective sector or entity.
Traditionally the majority of such funds were not available to the general public (i.e,
non-public offer), but this mix is changing. They charge members for services at
cost and fees, charges and insurance benefits are often subsidised by the sponsors or
through the fund surplus (APRA (2005)). In contrast, retail funds are public offer
funds that provide superannuation products to the general public on a commercial
’for-profit’ basis.
The most recent annual superannuation fund data for June 2012 is presented in
Table 2.1. This shows that SMSFs are the largest sector by total assets under
managements, with 31.5% of all superannuation assets in the Australian market.
Retail funds lead the institutional super fund market with 26.4% of market share,
while industry and corporate super funds trails at 19.1% and 4% respectively.
Table 2.1: Composition of the Australian Superannuation Industry 2012
Fund Type No. of Funds No. of Accounts Total Assets Average Balance
(AU$’000) (AU$ Billion) (AU$’000)
Corporate 122 551 56.1 101.8
Industry 56 11,664 267.3 22.9
Public Sector 39 3,371 222.7 66.1
Retail 135 15,408 371.4 24.1
SMSF (Small) 481,538 918 440.9 480.4
Other 42.2
Total 481,957 31,912 1,400.6 695.3
The dynamics of asset growth by the different fund types over the past decade is
illustrated in Figure 2.1.
10
Figure 2.1: Assets Under Management: Fund Type
Source: APRA Annual Superannuation Bulletin 2012
Despite the rapid growth of assets held in SMSFs, it remains a relatively small part
of the Australian Market in terms of members.3 The vast majority of Australian
workers (97.5%) still belong to the large institutional funds. Thus, the rest of
this section and subsequently the remainder of this research will only focu on the
institutional funds.
2.2.2 Coverage
The introduction of the mandatory Superannuation Guarantee saw rapid growth
in the coverage of superannuation schemes. Table 2.2 shows the decomposition of
superannuation coverage before and after the introduction of SG in 1992.
3Members of SMSFs are typically higher income earners and their families,
11
Table 2.2: Superannuation Coverage (1982 - 2012)
Year Full-time Coverage (%) Part-time Coverage(%) Total Coverage(%)
1986 46.5 7.0 39.4
1989 55.1 17.8 48.1
1992 88.0 54.1 80.3
2010 94.4 79.2 89.8
2012 94.1 80.2 89.9
Source: ABS No. 6310.0
Today, 89.9% of the Australian workforce are covered under the superannuation
system, which is the highest participation rate in private retirement saving among
comparable economies (OECD (2007)). The retirement and retirement intentions
survey conducted by the ABS indicate that in 2012 more than 43% of retired
Australians have benefited from superannuation at some time, of which 20% rely
solely on their superannuation lump sum payment to support their retirement.
Further, projections based on current trends also indicate that by 2050, only 28.3%
of retirees will entirely depend on the Age Pension to fund their retirement, private
savings in the form of superannuation are expected to partially or entirely fund the
rest of the retired population.4
2.2.3 Benefit Design
Traditionally, the majority of superannuation funds operate under a Defined Benefit
(DB) framework, where the retirement benefit of an individual is a function of certain
employment factors; benefit typically depends on the employee’s earning history, age,
tenure of service etc. In such schemes, the investment risk falls on the fund sponsors;
if the fund has inadequate assets to meet its obligations, the sponsor is required to
make up the shortfall. However, most DB plans suffer from poor portability, so
changes in employment would significantly reduce value of retirement benefit. The
4ABS Survey of Employment Arrangements, Retirement and Superannuation 2012.
12
limited portability of DB benefits creates a major disincentive against moving jobs.
This feature coupled with concerns about the ability of employers to maintain the
benefit structure during periods of market downturn (such as the GFC) saw Defined
Contribution (DC) plans rise in popularity. In the DC arrangement, investment risks
is borne entirely by fund members as well as the risk that inadequate contributions
are made into the plan. Under a DC scheme, the final benefit received by the member
is directly related to their SG contributions, their choice of investment strategy, the
performance of the chosen portfolio and the fees and cost charged by the fund.
Today, most DB schemes have completely or partially transitioned into DC
structures. That is, most DB schemes have now closed to new members Table
2.2 shows that the membership of DC plans has increased from 18.2% in 1983 to
94% in 2008.
Table 2.3: Membership Distribution: Benefit Type
Year Members in DB funds (%) Members in DC funds (%)
1982-83 81.8 18.2
1991-92 24.3 73.2
1999-00 13.9 86.1
2007-08 6.0 94.0
Source: Australian Social Trends, March 2009, ABS No. 4102.0
The increasing prevalence of DC schemes represents one of the most significant
changes in the Australian superannuation system. While mandatory superannuation
helps alleviate some of the fiscal responsibly of the government for retirement
incomes, the structural shift from DB to DC schemes places greater emphasis on
member’s choice of investment portfolios and the management of these portfolios by
superannuation funds, as their performance directly impacts the level of benefit
received and the ability of these private savings to deliver retirement income
requirements.
13
2.2.4 Regulatory Framework
With the exception of a small number of selected public sector funds which are
‘constitutionally protected’, all institutional funds in the Australian superannuation
industry fall under the jurisdiction of the Australian Prudential Regulation Author-
ity’s (APRA). APRA plays a key role in the industry as the administrator of The
Superannuation Industry (Supervision) Act 1993 (Cth) (SIS Act). The SIS Act
provides the authority with extensive supervisory and regulatory powers.
APRA’s regulatory approach is ‘principle-based’ rather than ‘rule-driven’. This
means it focuses on promoting prudent behaviour by superannuation trustees and
on ensuring the trustees comply with prudential requirements and operational
standards. The aim of prudential regulation is to promote best practices and
soundness within the funds, and reduce the likelihood that regulated funds will
fail and be unable to meet their commitments (Liu (2013)).
In practice, APRA relies on arm-twisting and moral persuasion to ensure compliance
(APRA (1998)). This passive regulatory stance typically translates to non-binding
guidance in the form of circulars and guidance notes to regulated superannuation
funds. However, the installation of a new prudential standard making power to
APRA in 2012 is likely to see potential regulatory reforms in the superannuation
sector, including stricter prudential standards, and more active regulatory involve-
ment as seen in other APRA regulated sectors such as Banking.
2.3 Industry Trends
The superannuation industry in Australia has undergone a tremendous and rapid
change over the past decade. This section will examine some key industry trends
that motivate this study of the efficiency of the superannuation industry.
14
2.3.1 Decreasing Levels of Competition
Over the past decade there has been a substantial consolidation of the number of
superannuation funds within the industry. Figure 2.2 illustrates the decrease in
the number of funds by 57.6% from 469 funds in 2004 to 199 funds in 2012.5 Over
this period, the Herfindahl Hirshman Index increased from 1.5% in 2004 to 2.8% in
2012, suggesting greater market concentration and decreasing level of competition
in the industry. Although, this figure remains relatively low compared to HHI
benchmarks,6 there is still a need to examine how this market rationalisation has
affected the efficiency of the industry and the welfare of members.
Figure 2.2: Market Concentration
.01
.015
.02
.025
.03
Her
finda
hl In
dex
100
200
300
400
500
Num
ber
of F
unds
2004 2006 2008 2010 2012Year
Industry Participants Herfindahl Index
Figure 2.3 shows that industry consolidation has not been uniform across public
offer and non-public offer superannuation funds: the number of public offer funds
decreased by 12.7% and non-public offer funds decreased by 80.1%.
5Figures are post sample selection, where erroneous observations and missing data are dropped.6Herfindahl Hirschman Index (HHI) indicates the level of concentration, and distribution of
market share, within an industry. An industry is typically considered low concentration when HHIis less than 15%.
15
Figure 2.3: Sector Decomposition
0
50
100
150
200
250
Num
ber
of F
unds
2004 2006 2008 2010 2012Years
Industry Funds Retail FundsCorporate Funds Public Sector Funds
The rationalisation within the non-public offer sector is primarily driven by market
exits of corporate super funds, while the number of Industry funds and Public
Sector funds have remained relatively stable. The number of corporate funds has
significantly decreased by 91.3% over the past decade. This is in line with the
observation that an increasing number of small corporate funds wind up and merged
their managed assets into retail master trust funds due to high cost associated with
onerous regulatory compliance and licensing requirements (from 2002). The number
of retail funds have also moderately decreased.
Current literature surrounding the superannuation industry in Australia has been
inadequate in assessing the impact of the rapid consolidation and asymmetric
rationalisation on overall industry efficiency. This research seeks to provide some
empirical evidence to inform such debates.
16
2.3.2 Stagnant Expenses
As discussed earlier, the industry has grown considerably in scale over the past
decade, and in conjunction with market consolidation, this has in effect resulted in
fewer and larger funds. Economies of scale imply that larger funds are more efficient
than smaller funds because they can distribute fixed costs across more members
and assets. However, Figure 2.4 illustrates that the superannuation industry has
been stagnant in improving average fees and charges imposed on members, with the
expense to assets ratio remaining at the 1% level.
Figure 2.4: Average Fund Asset and Expense Ratio
0
.005
.01
.015
.02
Exp
ense
Rat
io
0
1000
2000
3000
4000
Ave
rage
Ass
et (
$ m
illio
n)
2004 2006 2008 2010 2012Year
Average Assets Expense Ratio
This observation is somewhat consistent across the four institutional fund types.
Further, Figure 2.5 shows that public sector funds have the lowest reported fees,
immediately followed by corporate funds, industry and retail funds. This is
consistent with the prior assertion that these funds are likely to be subsidised by
their respective sponsors. Retail funds have the highest average expense ratio, again
not surprising as these funds are mostly ’for profit’ public offer funds.
17
Figure 2.5: Average Expense Ratio by Fund Type
.005
.01
.015
.02
Ave
rage
Exp
ense
Rat
io (
%)
2004 2006 2008 2010 2012Year
Corporate IndustryRetail Public Sector
18
Chapter 3
Literature Review
A comprehensive literature review of efficiency studies on financial institutions is
a fairly daunting task. There is no consensus among researchers as to the best
technique to measure efficiency (that is, should one use parametric approaches,
non-parametric approaches or semi parametric approaches), nor is there agreement
on the precise inputs and outputs to be calibrated in the analysis (in particular,
should one take intermediation approach or the production approach).
Further, the quantum of the efficiency and productivity literature is large. Therefore,
rather than canvassing every relevant study, this chapter will focus on the
key considerations one must make when undertaking a study of superannuation
efficiency, and illustrate them by citing relevant studies.
The literature review is organised as follows: Section 3.1 will briefly discuss the
key policies consideration presented by the Cooper Review and the motivations in
undertaking an efficiency study of the Australian superannuation industry. Section
3.2 canvas the most prominent area of efficiency study in financial services, that is
the U.S. banking market and mutual funds, and examine how these literatures have
helped to inform government policies and research interests. These studies deviate
from traditional productivity literature as the industry does not have conventional
production process, the sights from this stream of literature provides guidance
to calibrate our efficiency model for the superannuation industry. Section 3.3
extends our review of literature to more recent international studies on pension fund
efficiencies. And lastly, Section 3.4 discusses the exiting performance and efficiency
studies of the Australian superannuation industry, this section will also existing
identify research gaps and the potential contributions of this research.
19
3.1 The Cooper Review
In 2009, the Australian Government announced a comprehensive review of Aus-
tralia’s superannuation system: the Super System Review (the Cooper Review).
The review was chaired by Jeremy Cooper and was charged with ‘examining
and analysing the governance, efficiency, structure and operation of Australia’s
superannuation system. The Review is focused on achieving an outcome that is
in the best financial interests of members and which maximises retirement incomes
for Australians’ (super system review, p5). The Cooper Review was conducted in
three phases. In particular, phase two of the review sought to address whether
the superannuation system is operating as efficiently as it could be, whether the
full benefit of economies of scale are being harnessed and whether processes and
procedures that worked when the system was smaller are still appropriate.1
This section of the thesis give an overview of the Cooper Review and its key findings
as it applies to an assessment of superannuation fund efficiency in Australia. The
purpose is not to critique the review or its findings, but rather to identify the
potential sources inefficiencies in the Australian superannuation system as these
characteristics form the basis for our efficiency analysis.
The Cooper Review took the view that wholesale investment markets are competitive
and efficient, therefore only marginal potential gains in efficiency can be realised
by increasing gross investment outcomes. However, there appears to be scope for
improvement in the overall system efficiency by refining and streamlining operation
processes and reducing costs and leakages (including agency costs). In this respect,
the Cooper Review has identified several key areas of potential efficiency gains.
1Australian Superannuation System Review, Phase Two: Operation and Efficiency 2009
20
3.1.1 Reducing Costs
The first key area of efficiency gain is to reduce the universal fixed costs associated
with the industry so that service providers can reduce price of service, while still
retaining their current profit margin. The review noted that by removing redundant
regulatory requirements and promoting greater utilisation of new technologies in
super fund, one can reduce the relatively large fixed costs borne by current market
participants due to system design and external factors. Efficiency gains under this
regime is likely to occur if the market is highly competitive and that any potential
gains for super funds is likely to be fully passed on to consumer in the form of
reduced service fees.
The Cooper Review also recognised cost transparency as a major impediment to
achieving greater efficiency in this respect. In particular, the emphasis on lowering
prices would likely result in too much focus on services such as administration and
insurance where fees are more transparent, and insufficient attention on achieving
efficiencies from areas such as investment management and distribution, where the
costs imposed on members are less visible.
3.1.2 Lower profit margins
The second key area of efficiency gain is to promote vigorous price competition in
the market where it is insufficient, particularly in areas where price competition
between service providers are less than optimal. The Cooper Review argued that in
selecting professional service providers, super fund trustees should be more aware
of the profitability of their service providers. The current lack of such information
limits the extend of competition between service providers and inhibits the ability
of trustees to make fully informed decisions in pursuing the best financial interest
of their members.
The findings of the Wallis inquiry (1997) into the Australian banking sector are
21
also relevant. The final report suggested that an increase in competition for
(profitable) aspects of the financial services market by specialist providers may
result in a reduction of inefficient product pricing (cross-subsidisation) by existing
firms. For example, the administrators of superannuation funds could be inclined
to provide transaction services at a discounted price in order to obtain patronage
of customers (super fund members), the loss of revenue from transaction services
could be then made up in other aspects of service such as investment and insurance
charges. An increase in competition due to greater profitability transparency in
the superannuation investment service market could reduce the service provider’s
ability to recover the foregone revenue from their transaction discounts. As a result,
one might expect the cross-subsidisation between services to be removed, hence the
overall result of increases in competition may not be optimal.
3.1.3 Technology
Coupled with changing demographics is the way in which members use technology
to interact with their superannuation funds. Information is now increasingly
susceptible to transmission, storage and interrogation because of advances in
technology and reduced cost barriers. While members have increasing access to
information to make decisions, super funds also have an increasing responsibly to
use and interrogate the information they possess. Super funds which operates in an
competitive environment would have to justify their value where customers have,
because of technology, direct access to markets.
The Cooper Review argued that the range of technologies employed by different
participants in the super system, the existence of multiple legacy technologies within
some organisations and the retention of manual and paper based processes in other
areas, seem to present challenges to the efficient operation of the system.
22
3.1.4 Default Portfolios
The Cooper Review cited as at June 2008, 46% of total assets in APRA-regulated
funds were in default investment options, ranging from 23.4% of assets in retail
funds to 74.8% in industry funds (the APRA data for June 2012 suggests a slight
reduction since then). This proportion of assets translates to a higher proportion of
members, as members with lower account balances are over represented in default
investment options.
Further, most default investment options in super funds are not age based, and
typically use a ’balanced’ asset allocation.2It is then important that super funds have
default investment options that appropriately accommodate a variety of members
and that members’ best financial interests are served.3
3.1.5 Regulation
The Cooper Review posed the question of whether the cost effectiveness and
usefulness of regulation in superannuation can be assessed? That is, how can
regulation be frequently reviewed?
Shared administration of the same provisions by different regulators, with their
distinctive approaches to fulfilling their statutory mandates, may lead to particular
problems as industry sectors diverge in different directions. It is then imperative
that the regulators coordinate efficiently and do not impose excess compliance and
reporting obligations on the industry. Further, there is a need to ensure that
regulatory standards are sensible and promote efficiency in delivering best outcomes
for members.
2In Australia, ’balanced’ most commonly refers to an approximately 70/30 growth/defensiveasset allocation. By contrast, some other countries typically use far more conservative defaultinvestment options for their default fund.
3The new legislation setting of the new MySuper product endorsed age based defaults.
23
3.2 International Literature
3.2.1 Efficiency Studies: General Applications
Berger and Humphrey (1997) surveyed 130 efficiency studies, the majority in the
banking sector. This survey revealed that the literature has primarily contributed
to three principal areas: informing government policy; addressing research issues;
and improving managerial performance. This classification and its subcategories are
useful in defining the key focus of this study and assessing its potential contributions.
The classification by Berger and Humphrey (1997) is reproduced in Table 3.1. This
section will briefly discuss each research area and particularly where it relates to the
core analysis and contributions of this research.
Table 3.1: Research Areas - Berger and Humphrey (1997)
Rsearch Areas Subcategories
Inform government policy Deregulation, financial disruption
Institutional failure, risk, problem loans
Market structure and concentration
Mergers and acquisitions
Address research issues Profit, revenue
Confidence intervals
Comparing different efficiency techniques
Different output measures
Organisational form, corporate control issues
Cross-country comparisons
Methodology issues
Opportunity cost, output diversification
Improve managerial performance Credit unions
Bank branch
Savings and loans branch
Informing Government Policy
Deregulation - One of the main sources of potential efficiency gain, identified by
the Cooper Review, was the deregulation of the superannuation market in Australia.
Indeed, financial deregulation has often been characteristically undertaken to
24
improve the efficient performance of an industry, whereby a reduction in redundant
regulatory requirements could promote better resource allocation and greater
competition in the industry. This could then lead to a reduction in prices and
the delivery of better services to consumers.
Of the studies canvassed by Berger and Humphrey (1997), there was no clear
evidence that deregulation results in increased efficiency. For example, in the U.S.,
measured efficiency fell following deregulation. Bank productivity declined largely
due to interest rate deregulation which resulted in higher interest rates being paid
on consumer deposits, with no accompanying reduction in the services provided to
those account holders and no increase in account fees. While some jurisdictions
shared similar experiences to that of the U.S., selected studies did show evidence
of increases in efficiency following financial deregulation of the sector. Berger and
Humphrey (1997) concluded that the effects of deregulation depends, in part, on
the initial industry conditions and therefore ‘the conventional wisdom which holds
that deregulation always improves efficiency and productivity may be incorrect’.
Market structure and concentration - Berger and Humphrey (1997) noted that
many studies of financial institutions and other firms have found a positive statistical
relationship between market concentration and profitability. This observation has
often been attributed to market powers, in which firms can earn supernormal
profits. Alternatively, due to more efficient firms gaining increased market share,
when combined with low production costs result in increased industry profits.
These competing arguments of profitability, market power vs. efficiency, have
opposing implications for the government’s anti-competition policy. If high profits
are fashioned by market power, anti-competitive actions are likely to be socially
beneficial. If high relative efficiency is the grounds for greater profitability, then
breaking up efficient firms, which have gained large market shares, or by blocking
mergers and acquisition of firms, are likely to be less favourable to consumers.
25
The reality of the superannuation industry in Australia, is an industry undergoing
rapid consolidation and increase in market concentration. This is particularly so
for the public offer funds, where corporate super funds commonly merge into retail
master trust funds, and the consolidation in the retail superannuation market itself.
It is then imperative to understand the fundamental underpinning of the competition
policy proposed by the Cooper review.
Addressing Research Issues
Organisational Form and Corporate Control - Berger et al. (1993) identified
several possible determinants of inefficiency, including agency problems between
owners and managers, regulation, organisational and legal structures. In theory,
firms owned by shareholders are expected to be more efficient due to the incentives
asserted onto management to control costs and enhance profits. Berger and Mester
(1997) provide empirical evidence for this proposition, where publicly listed banks
as well as those owned by a parent entity are confirmed to be more efficient.
Parallels can be drawn between the governance structure of the banking sector and
the Australian superannuation industry, where each fund type (Industry, Retail,
Public Sector, Corporate) have distinctive governance features. For example, ’for-
profit’ super funds operate under similar management incentive schemes to those
of the publicly listed banks, and corporate funds are typically governed by its
sponsoring entity. Additional board features could also contribute to the efficiency
of the fund.
Methodological Issues - Literature in this category focuses on improving the
technique by which efficiency is estimated. Advances has been made in areas
associated with semi-parametric and semi-nonparametric techniques which depart
from traditional DEA/SFA analysis. However, these more advanced techniques are
26
out of the scope of this thesis.
An interesting technique considered in selected studies using data with time
dimensions is the Malmquist Index. The technique allows researchers to account
for technological change in analysing the efficiency of firms.4. This is outside the
scope of this research, and could be pursued in future studies.
Wheelock and Wilson (1999) studied changes in cost productivity for over 11,387
U.S. banks between 1984 and 1993 using DEA constructed Malmquist indices. To
analyse the sources of change in productivity, the researchers decomposed the decline
in productivity into four components: changes in pure technical efficiency, changes
in scale efficiency, changes in pure technology and finally, changes in the scale of
technology, as follows.
Productivity =∆Efficiency× (∆Technology)12
=(∆Pure Technical Efficiency× Scale Efficiency)
×∆Pure Technology12 × Scale Technology
12
The change in pure technology is the geometric mean of a ratio that measures the
shift in variable returns to scale (VRS) frontier relative to the firms’ position at time
t+1, and a second ratio, which measure the shift in the VRS frontier relative to the
firms position at time t. Change in the scale of technology describes the change in
returns to scale resulting from technological changes.
Using U.S. bank data from 1984 to 1993, Wheelock and Wilson (1999) found large
increases in the change of pure technology, which indicates that the efficiency
frontier shifted heavily outward during the period. The paper also found overall
cost efficiency declined between 11.43% to 17.28%, depending on the bank size
(smaller banks have larger declines than larger banks). This suggest that despite
4Recall that the ability of super funds to adapt to technological shifts in the market wasidentified as a key efficiency metric by the Cooper review
27
large advances in technology during the period, average technical and scale efficiency
has declined more than the opposing movements of the frontier. The overall effect
is a decline in productivity, which is less for large banks than for small banks.
3.2.2 Efficiency Studies: Pension Funds
The application of efficiency analysis to pension or superannuation funds is limited,
and is somewhat concentrated to a few South American countries. To our benefit,
a number of these countries followed Chile in privatising their pension systems,
and the resulting pension fund landscape is somewhat similar to that in Australia.
Therefore, it is useful in canvassing some of the literature in this space and their
experiences in assessing the efficiency of pension funds.
Earlier work of Braberman et al. (1999) studied Argentinian pension fund manage-
ment institutions using a Cobb-Douglas cost frontier model,5 where the researchers
regressed operating cost again three independent variables: the number of members;
the positive turnovers; and the profitability of the fund. The study also included two
dummy variables to account for changes in regulation after November 1997. They
found that regulation increased total costs but did not significantly affect relative
efficiency, although the study was criticised for its misspecification of the Cobb-
Douglas production function due to the lack of factor prices.
Barrientos and Boussofiane (2005) analysed Chilean pension fund management
companies with the use of a two-stage procedure. The researchers first computed
efficiency scores for the pension funds using the non-parametric frontier approach
Data Envelopment Analysis (DEA), and then regressed the scores against contextual
variables using OLS regression. The study was conservative in defining the output
factors of pension funds. Similar definitions as those used in the banking sector
were followed, including total revenue and the number of contributors as outputs;
5This is one possible specification of Stochastic Frontier Analysis (SFA), which will be discussedlater in this thesis in conjunction with Data Envelopment Analysis.
28
marketing costs, office personnel costs and administration costs as inputs. The study
found no continuous trend toward an improvement in technical efficiency. Analysis
of the determinants of efficiency showed that an increase in market share contributes
positively to technical efficiency. Barros and Garcia (2006) noted that the second
stage analysis was the caveat of this paper, where the regression model did not
fully account for the unique nature of the efficiency scores (continuous on the unit
interval). Further, studies by Simar and Wilson (1999, 2000) showed the efficiency
scores obtained in the first stage are correlated with the explanatory variables used
in the second stage, therefore any regression output is likely to be inconsistent and
biased. Barros and Garcia (2006) suggested that a bootstrap procedure is necessary
to overcome this problem.
Using the same data, Barros and Garcia (2006) studied the sample with four DEA
models and concluded that traditional DEA models are unable to discriminate
adequately between Portuguese pension funds with most funds on the VRS frontier.
The study also found significant statistical correlations between efficiency and
selected contextual variables, this include private/public status, mergers/acquisition
and scale. However, the study did not conduct any second stage regression analyses.
Further work by Barros et al. (2008) analysed the sample using Stochastic Frontier
Analysis (SFA) and found that market share is negatively related to cost efficiencies,
while mergers and acquisitions appear to increase cost efficiencies.
In addition to the above attempts at analysing of pension funds, there is also
some evidence in the mutual fund context. Basso and Funari (2001) constructed
a generalised performance index for Italian mutual funds using DEA, where the
researchers defined an innovative output measure which reflects the matching of
the investor’s preference structure and the time occurrence of the returns. This
study departs from traditional efficiency analysis in focusing on either cost or profit
definitions of outputs, and was among the first to include variables that directly
29
capture specialised functionality of the industry studied. Further studies by Basso
and Funari (2003) builds upon this aspect by introducing ethical indicators as a
key output for mutual funds, the measure not only allows mutual funds to compete
on financial performance but also on their ethical standards, such as investment in
renewable energy and community projects.
Galagedera and Silvapulle (2002) used DEA to analyse the relative efficiency of
257 Australian mutual funds, while secondary regression explored the relationship
between fund attributes and their respective efficiency measures. The study found
that technical efficiency of mutual funds in Australia appear to be dominated by the
effects of scale efficiency compared to pure technical efficiency. The study found no
significant relationship between the mutual funds’ relative efficiency score and its
structure, classification, size and age.
3.3 Australian Literature
Efficiency studies of the Australian superannuation industry are sparse with only
one relevant frontier efficiency analysis conducted. Sathye (2011) examined the
efficiency of retail superannuation funds in Australia for the period 2005 - 2009 using
DEA. The study found a positive relationship between fund size and production
efficiency, and suggested that M&A within the industry should be encouraged to
achieve higher overall efficiency through rationalisation. The caveats of this study is
in two fold. First, the study did not consider other fund types (Industry, Corporate,
Public Sector) which accounts for a significant proportion of the Australian market.
Further, output variables defined in the study were poor in reflecting the features
of the superannuation industry in Australia. Second, The study was inadequate in
accounting for the prevalence of alternative second stage techniques popularised in
the recent DEA literature. TOBIT regression employed in this study was widely
recognised as a misspecification of the DEA efficiency scores.
30
Other industry studies in Australia have been focused on particular elements of super
funds’ performance indicators, including investment returns, risk, administrative fees
and fund designs. This research aims to extend these studies by adopting efficiency
analysis techniques used extensively in other financial services industries. The study
is valuable in that it takes a broader view of super fund operations and assesses its
performance based on a combination of its key functions other than investment
outcomes.
3.4 Investment Performances and Risk
Earlier studies often employed traditional financial indicators commonly used in the
analysis of other professionally managed funds. Bird et al. (1983) examined the
performance of Australia superannuation funds from 1973 to 1981 using the Sharpe
ratio, the Treynor ratio and Jensen’s alpha. The study compared funds’ investment
performance with three standardised benchmark market portfolios. They found no
consistency in funds performance but did note that funds with larger asset pools
performed relatively better than smaller funds. This was first amongst an extensive
body of inquiry that examine the performance of asset managers using risk-adjusted
returns.
Jacob (1998) argues that, because a benchmark portfolio is intended to represent
a passive alternative to the investment manager’s style, then if there are costs
associated with replicating this passive portfolio, those should be reflected in the
benchmark’s return, or the comparison is not fair. Drew and Stanford (2001) took
this into account by constructing a customised indexed portfolio for trustees, based
on the notion of relative market efficiency. The portfolio bears the properties of an
open end retail fund, and the management fees are proxied from US mutual funds.
The study found that low cost passive asset selection provides superior risk adjusted
returns to fund members than the returns achieved through a high cost active asset
selection strategy. The study is further extended in Drew and Stanford (2003),
31
where the authors used a multi factor model to analyse the performance of retail
superannuation funds in Australia. The study found active investment strategies did
not add any value for members, instead active portfolio management consistently
under performs passive portfolios. Due to the limitation of data, only the retail
segment of the industry was analysed. In addition, the analysis relied heavily on
benchmark portfolios; this meant the studies were highly sensitive to assumptions
made in regards to portfolio asset allocation and management fees.
Coleman et al. (2006) conducted a broader industry study using two sets of
proprietary APRA data. The study examined the relationship between risk, returns
and expenses over the period 1995 to 2002 for large APRA funds, with a particular
focus on how performance varies by total assets, by fund type and over time. By
comparing net return on assets and volatilities across fund types, the study found
that retail and industry funds have the lowest returns and volatilities whereas public
sector and corporate funds have the highest returns and volatilities. On both a net
return and risk adjusted return basis, not-for-profit funds significantly outperformed
for-profit funds. The differences are attributed to the greater agency costs of
for-profit funds (due to non-representative trustee board structure and potential
board member conflicts of interest). Furthermore, the study also found a positive
relationship between returns and fund size.
Ellis et al. (2008) recognised that prior studies of fund performances were limited
in terms of comparability between funds, due to active portfolio choice of members;
thereby implicitly incorporating asset preferences of its members in its performance.
To focus on comparable returns, the study focuses on the default option only and
uses the asset allocation information to construct a benchmark for the default option
in each fund. Comparisons of performance are then made relative to this benchmark.
This technique is superior to prior studies as it isolates the net performance that
a fund member would earn using the trustee’s asset allocation and management
32
approach. The data used in this paper is a customized quarterly survey of 197
large superannuation funds between 2001 and 2006. The result showed significantly
lower average net return relative to the benchmark for balanced and growth retail
default investment options, compared with other fund types. This implies both
higher expenses and taxes, explicit and embedded, for the retail fund options.
However, the study is limited in that it only considered the risk element in relation
to the benchmark portfolio but not the fund’s portfolio. This shortcoming could be
attributed to the limited time series of the dataset.
So far, the literature that assesses Superannuation fund performance has been based
on risk to reward comparisons, though they differ in procedure. The definitions of
risk often rely on theoretical assumptions and may be open to manipulation. Sy and
Liu (2009) took a different approach to the risk measures. This study argues that
traditional ratios such as the Sharpe ratio is inappropriate in situations where the
manager has control over the volatility or is encourage to assume different levels of
risk from the benchmark. To illustrate, the paper proposed the following example:
Table 3.2: Negative Sharpe Ratio Example - Sy and Liu (2009)
rf = 5% Return Volatility Sharpe Ratio
(%) (%) (%)
Benchmark -2 10 -0.7
Manager A -5 20 -0.5
Manager B 1 5 -0.8
Even though Manager A underperformed its benchmark index and assumed
considerably more risk, it still out-performed both the benchmark and manager B in
terms of the Sharpe ratio. Scholz and Wilkens (2008) described this phenomenon as
’market climate bias’ where a fund with constant fund specific characteristics that
outperforms the Sharpe ratio of the market index in a declining market will not
necessarily have a superior Sharpe ratio in a normal market period. This paradox
originates from the fact that the Sharpe ratio is a mere measure of efficiency of
33
risk taking: how much excess return per unit of risk taken. It does not reward or
measure the decision of the manager to select the appropriate level of risk.
Given the issues with existing risk to reward ratio measures, Scholz and Wilkens
(2008) proposes an alternative approach called risk-adjusted value added (RAVA)
metric.
ρ =RA −RB
σB
Where RA is the total fund return of the firm, RB is the benchmark return, B
is the benchmark volatility. The measure is advantageous as it departs from the
over emphasis on return volatility and avoids the contradiction experienced by
the Sharpe Ratio. Further the metric can also be computed with relative ease,
as data availability for the benchmark is much more abundant. Using this index
and total fund level accounting data, the study found a statistically significant
inverse correlation between the net performance metric values and management
expenses of the firms. That is, on average, value adding from active management
appears statistically to be unable to overcome higher costs associated with attempts
to exploit market inefficiencies.
3.5 Administrative Expenses
Another key area of interest revolves around the effectiveness of funds’ operations,
particular in regards to the administrative fees charged by superannuation funds.
The importance of administrative fees is highlighted by Bateman (2002), which
noted that costs arising from custodianship of funds could have a profound impact
on the financial well-being of retirees. Individual retirement accumulations could be
reduced by as much as 22% for a 1% per annum management fee. In conjunction with
contribution and superannuation taxes, a 1% p.a administrative fee could reduce the
then statutory contribution rate of 9% to an effective rate of 5.1%.
34
Bateman et al. (2001) reported that the costs within the Australia superannuation
industry for the financial year 1998-1999 averaged 1.72%. This paper also raised
concerns in regards to the considerable variation of fees between types and
characteristics of superannuation funds. Most notably, retail funds exhibits 50%
higher administrative charges than industry average.
Other studies, including Drew and Stanford (2003) recognise the persistent high
level of administration costs of superannuation funds in Australia. The authors
attributed these costs and other member benefit issues to the agency issues resulting
from constraints on market competition. The study advocates unrestricted member
choice of fund and portability of superannuation entitlements, and argues that funds
can be disciplined by increasing market competition; provide incentives to improve
operating efficiency.
In contrast, Vidler (2004) examined optimal market structure and makes the
observation that such an approach ignores cost issues inherent in personal account
based pension systems, especially with non-centralised account administration. It
also ignores the added costs associated with competition. The paper supports
the argument for achieving a reasonable level of fees within the superannuation
industry, however asserts that free choice would compound rather than alleviate
current problems related to fees. The paper alternatively advocates for an
institutional market structure, similar to that of the Swedish pension system.
Account administration in that market is public and centralised, preventing account
proliferation and maximising administrative economies of scale. Consumers choose
from a relatively limited range of funds and investment options. Fund management
is conducted by private firms which periodically tender for licenses at auctions run by
the government. This competition ’for the market’ rather than ’within the market’
provides more effective discipline on providers.
35
Bateman and Mitchell (2001) took a more structural approach. The studies focuses
on the design features of superannuation fund structure and how it could best deliver
value for its members. In this respect, the paper examined the linkage between fund
designs and their respective administration costs. Using data from 1998-1999, the
study reported that DC plans would likely to bear the least amount of administration
cost; this is followed by DB plans and retail funds. There also appears to be a
significant negative relationship between the fund’s scale and its respective costs.
36
Chapter 4
Theoretical Framework
This chapter is concerned with the theory underpinning efficiency and productivity
measurement. Section 4.1 briefly summarises the underlying theory of frontier
efficiency analysis. Section 4.2 discusses two popular techniques in estimating
frontier efficiency and investigates the considerations in selecting the correct
approach within the context of this research.
4.1 Efficiency Analysis
4.1.1 Farrell Efficiency
Frontier analysis is essentially a sophisticated way to ’benchmark’ the relative
performance of firms (super funds). Farrell (1957) was among the first to
conceptualise this framework, by developing a technique that could measure the
productivity and efficiency of firms using multiple input and output factors. Farrel’s
research departed from previous literature, which utilised simple labour productivity
as a measurement of firm efficiency, by incorporating other essential inputs including
capitals. The technique also accommodates and avoids the index number problems.
Through the analysis of frontier dynamics, Farrell was able to isolate the two
components of firm efficiency measures, technical and allocative efficiency. Technical
Efficiency (TE) measures the ability of a firm to achieve maximum output given the
set of input parameters, and Allocative Efficiency (AE), reflects the ability of a firm
to use inputs in optimal proportions, given their respective prices. The firm’s Total
Overall Efficiency (TOE) is then the product of the two elements.
q = F (x1, x2)
Where Q is the feasible vector of outputs given input vector x1 and x2. By assuming
constant return to scale, or homogeneous of degree zero; we are able to present all
37
the relevant information in a simple ’isoquant’ diagram in two dimensional space.
Figure 4.1: Input-oriented efficiency measurement and the isoquant
Source: Coelli et al. (2005), p165
The ’isoquant’ SS’ represents the various combinations of the two factors that a
perfectly efficient firm might use to produce unit output. It is also the frontier that
represents the minimal amount of inputs required to produce one unit of output;
now the point B’ represents efficiency firm using the two factors in the same ratio
as B. It can be seen that it produces the same output as B using only a fraction
OB’/OB as much of each factor. Farrell defined this ratio as Technical Efficiency of
firm P, and it could also be expressed in terms of input distance function di(x, y)
TE =1
di(x, y)
The firm under consideration is technically efficient if it is on the frontier, in which
case TE=1 and di(x, y) also equals to 1.
By the same token, if we observe price information w of the firm’s production factors,
38
we are able to construct an ’isocost’ line of the firm. The slope of such line KK’ is
simply the price ratio of the two inputs. This then allows us to measure the extent
to which a firm uses the various factors of production in the best proportions or
the allocative efficiency of the firm. Such measure allows one to further distinguish
performance among technically efficient firms, for example, of all the technically
efficient firms operating on the line SS’, firms operating at point C have the optimal
method of production rather than at B’. The costs of product at C will only be a
fraction OC/OB’ of those at B’, this also represent the possible reduction of costs
in its factors of production (Bhagavath, 2006).
Formally, if the observed firm B were perfectly efficient, both technically and in
its allocation of production factors, its costs would be a fraction OR/OB of what
they are. This Total Operating Efficiency of the firm could also be expressed as the
product of its technical and allocative efficiency. As can be seen below.
TOE =OR
OB=OB′OB
× OR
OB′
Two central problems to Farrell’s analysis of firm efficiency revolve around the
assumptions of constant returns to scale and the convexity of the isoquant line.
The latter is common among other economic theories; the assumption of convexity
ensures that the weighted average of any two attainable points is also attainable.
This then allows the construction of a hypothetical firm as a weighted average of
two observed firms, in the sense that each of its inputs and outputs is the same
weighted average of those of the observed firms; the weights being chosen so as to
give desired factor proportions (Farrell, 1957).
4.1.2 Scale Efficiency
The assumption of Constant Return to Scale (CRS) allows for computational
simplicity of the efficient set but imposes global scale efficiency on all firms. This
is unlikely to hold in practice. It is possible that a firm is both technically and
39
allocative efficient but the scale of operation of the firm may not be optimal.
Possible sources of scale inefficiency could be attributed to imperfect competition,
government regulations, constraints on finance etc. Suppose the firm involved may
be too small in its scale of operations, which might fall within the increasing returns
to scale (IRS) part of the production. Similarly, a firm may be too large and it may
operate within the decreasing returns to scale (DRS) part of the production function.
In both of these cases, efficiency of the firms might be improved by changing their
scale of operations . Farrell (1957) devised a methodology that divide firms in a
discrete output interval, the assumption being that returns are constant within a
group to a sufficient degree of approximation. However a key shortfall of this method
is that it does not isolate the effects of scale efficiency and also requires a sufficient
number of observations in each output interval which is not readily available. As a
result comparability between firms across output intervals is limited.
Given the radical shifts in the scale of the superannuation industry in Australia,
isolating the effect of scale efficiency could be the key to facilitate the explanation
of productivity changes of the industry over time.
Balk (2001) provides a formal framework to define scale efficiency and to study the
role of scale efficiency in productivity change. Consider the following one input and
single output case:
40
Figure 4.2: The effect of Scale Efficiency
Source: Coelli et al. (2005), p59
Where A,B,C are all technically efficient firms since they lie on the production
frontier. However, because the productivity of each of these firms is equal to the
ratio of their observed output and input quantities (i.e y/x), and this is equivalent to
the gradient of a line drawn from the origin through the data point (x,y), we can see
that firm B is more productive than others (highest gradient). Firm A is operating
in the increasing returns to scale portion of the production frontier. That is, it could
become more productive by increasing its scale of operation towards point B. By
the same token, firm C is in the decreasing return to scale portion and can improve
productivity by reducing its scales towards B.
The line representing firm B is tangential to the VRS production frontier, which
implies that it is operating at the highest feasible productivity. Most literature
refers to this point as the technically optimal productive scale (TOPS). The point
can be formalised as:
TOPS = max{yx
∣∣∣ (x, y) ∈ S}
41
Recall that under the assumption of CRS, all firms operates under perfect scale
efficiency, therefore the line through TOPS is also the CRS production frontier.
Under this simple model, one can decompose scale efficiency as the ratio of technical
efficiencies under CRS and VRS.
Figure 4.3: Scale Efficiency
Source: Coelli et al. (2005), p61
The technical efficiency of firm D relates to the distance from observed data point
to the VRS technology and is equal to the ratio:
TEV RS =GE
GD
Similarly, technical efficiency under CRS could be expressed as:
TECRS =GF
GD
The requires scale efficiency of firm D relates to the distance from the technically
42
efficiency data point,E, to the CRS technology:
SE =GF
GE
Simple manipulation reveals that this ratio could also be represented as the ratio
between the two technical efficiency measures:
SE =TECRSTEV RS
=GF
GD· GDGE
=GF
GE
Fare et al. (1998) has generalised this approach to multiple output/input factors to
the following form.
SE (x, q) =di (x, q | V RS)
di (x, q | CRS)=TECRSTEV RS
This thesis will following this framework in isolating the scale efficiency effects.
4.2 Efficiency Score Estimation
4.2.1 Model Selection
Theoretical best practice firms typically cannot be calculated from observed data,
which makes the efficiency concepts difficult to operationalise. Early attempts
used OLS regression technique that fits an average curve through the sample and
compares firms with the average performance in the industry. This technique is
not satisfactory as it does not produce the required efficiency measurements. This
has led to attempts to approximate best practice firms in the sample by estimating
frontiers.
Two most prominent frontier approaches are Data Envelopment Analysis and
(DEA) and Stochastic Frontier Analysis (SFA). DEA is a deterministic means of
constructing a ’piece-wise linear’ approximation of the efficiency frontier, where the
resulting frontier is ’kinked’ curve that ’envelops’ the sample, each observation is
43
optimised individual in reference to the frontier (Charnes et al., 1978).
SFA is an alternative approach to the traditional regression technique, where a
secondary error term is introduced to capture the impact of inefficiencies on the
production function while the traditional error captures the disturbances. This
technique could account for outliers which either are very atypical or appear to
be exceptional performers as a result of data measurement error. It is also useful
in capturing abnormal disturbances in firm performance, especially in sectors that
have highly volatile outputs and are prone to heterogeneous shocks. In contrast, the
inability of DEA to accommodate for these disturbances is constantly criticised in
the relevant literature, where the efficiency estimates could potentially be biased by
idiosyncratic shocks to its outputs.
However, SFA implicitly requires underlying assumptions regarding the production
function; the production process defines the deterministic component in the SFA
regression. This characteristic makes the application of SFA in non-traditional
productive industry such as the superannuation or pension industry extremely
difficult. This is because the true production function is not readily observable,
thus SFA application in such a scenario is likely to be vulnerable to specification
error without further modifications to the model (Grosskopt, 1993). In contrast,
DEA does make such assumptions, thus are more superior to SFA in this respect.
A clear advantage of DEA is the ability to readily incorporate multiple inputs and
outputs; whereas SFA is somewhat limited to singular output. In the context of this
thesis, our a priori knowledge of the superannuation industry in Australia is that it
is a crucial part of our social infrastructure and thus serves several key functions. As
a result the ability to accommodate for multiple vectors of output measures would
be more appropriate in capturing the entirety of the industry’s operations. Further,
DEA only requires information on the output and input quantities instead of its
44
prices (as required under SFA). This makes DEA particularly suitable for analysing
the efficiency of financial service providers, where it is difficult assign prices to many
of the outputs and input factors.
It is also possible to determine the underlying sources of inefficiency under DEA, as
the technique provides a means of decomposing economic inefficiency into technical
and allocative inefficiencies. Furthermore, DEA also allows technical inefficiencies
to be decomposed into scale effects, the effects of undisposable inputs, and a
residual pure inefficiency component. The decomposition of technical inefficiencies
could provide valuable insights for individual case studies of decision units (i.e.
superannuation funds) and could allow the industry regulator (i.e. APRA) to readily
identify sources of inefficiency in the market.
4.2.2 Limitations of DEA
Standard DEA models typically impose monotonicity (i.e. free decomposability of all
inputs and outputs) and convexity assumption. These assumption can sometimes be
overly restrictive and easily violated in practice. Congestion of production factors
and increasing marginal product of inputs are common examples of violations in
these assumptions. This will discuss further within the context of our analysis later
in this paper.
Due to the deterministic nature of DEA, one must also be cautiously aware of the
sensitivity of DEA to measurement errors in the data. Outliers in output and input
factors can significantly distort the shape of the frontier and reduce the efficiency
scores of nearby decision units. Therefore a vigorous outlier detection and sample
selection process is undertaken in constructing the sample used in this thesis.
In addition, sample size also has a significant impact on DEA results. Increasing
the sample size will tend to reduce the average efficiency score, because including
more firms provides greater scope for DEA to find similar comparison partners, and
45
thus reducing the dimension in which a firm could be relatively unique. Relevant
literature that investigate the required sample size for valid DEA analysis is explored
in later sections.
Further, DEA scores are also extremely sensitive to the selection of output and input
vectors. Previous studies have shown that subtle changes in parameter specifications
of DEA can drastically alter the efficiency estimates. The majority of the DEA
studies define output/input parameters in accordance with the assumptions about
the key functionalities of the entities being studied. In this research we adopt this
approach and discuss the merits of various output and input measures analysed by
existing industry literatures.
It is also important to note that DEA results are very limited in their comparability
outside the sample. Efficiency scores are only relative to the best practice firm within
the particular sample. Thus, it is not meaningful to compare the scores between
two different studies because differences in best practice between the samples are
unknown. Similarly, a DEA study that only includes observations from within
the nation cannot tell us how those observations compare with international best
practice firms. Although it is not the aim of this thesis to draw international
comparisons, this does limit our ability to draw parallels from similar studies in
other industries or countries.
Furthermore, the ’piece-wise’ comparison approach used in DEA enables peer to
peer comparison. This provides a powerful benchmarking capability where efficient
role models could be separately identified in different operating environments. Such
benchmark could be useful in guiding improvement of operations for inefficient
entities.
The focus of this research is to assess the performance the Australian superannuation
46
industry and to inform regulatory reforms. In this respect, DEA is the most
appropriate technique on account of its ability to capture the multiple service
functions that the industry provide to its members, as well as its ability to yield
valuable benchmarks and efficiency decompositions.
The comparison between the two models in relation to their ability to accommodate
stochastic shocks is less clear. Intuitively, SFA is superior in the sense that it
does not punish firms as much for performance shocks that are beyond the firm’s
control. However given the environment in which superannuation fund operate in,
it is unlikely that subgroups of funds will experience heterogeneous shocks. As a
result it is relatively benign to make the assumption that individual participants
in the industry faces fairly uniform changes in market conditions and economic
disturbances. That being said, it is possible that super funds experience fund
level externalities, therefore we check the robustness of our result using aggregated
performance over the sampling period.
4.2.3 Data Envelopment Analysis
Data Envelopment Analysis is the term first introduced in the operations research
literature by Charnes et al. (1978) (CCR) to measure the technical efficiency of a
given observed decision making unit. The technique is a non-parametric approach
that places relatively little structure on the specification of the production frontier.
The original model by CCR was restrictive as it assumes constant return to scale.
Their linear programming formulation adopted the piece wise linear convex hull
approach to frontier estimation, proposed by Farrell (1957). The technique also
accommodated for multiple inputs and multiple outputs. Banker et al. (1984) (BCC)
extended the CCR model to allow variable returns to scale and showed that solutions
to both CCR and BCC allowed a decomposition of CCR efficiency into technical
and scale components.
This section will explain the theoretical foundations of the CCR efficiency estimation
47
under constant return to scale and its VRS extension under BCC.
Constant Return to Scale (CRS)
Coelli et al. (2005) introduced DEA via the ratio form. For each firm, we would like
to obtain a measure of the ratio of all outputs over all inputs, such as u′yiv′xi
, where
u is an M × 1 vector of outputs weights and v is a N × 1 vector of input weights.
Q and X are matrices of output and inputs respectively. The optimal weights are
obtained by solving the mathematical programming problem:
maxu,v
u′yiv′xi
s.tu′yjv′xj
≤ 1, j = 1, 2, . . . , I,
u, v ≥ 0
The system of equations will calibrate the value of global weights on inputs (v)
and outputs (u) that maximises the efficiency score of sample firms, subject to the
efficiency scores being less than 1 and weights are greater than 0. However, this ratio
formulation has an infinite number of solutions, since any linear combination between
the weights and a common scalar is also a solution to the system of equations. To
avoid this, one can impose the constraint v′xi = 1, which provides the following
linear programming system of equations:
hj = maxu,v
µ′yi,
st v′xi = 1
u′yj − v′xj ≤ 0, j = 1, 2, . . . , I,
u, v ≥ 0
This form of linear programming problem of DEA is known as the multiplier form.
This model gives a value based measure of efficiency of DMU j. The weighting
48
variable u can be seen as an imputed marginal value of output y. Similarly, v can be
seen as the imputed marginal value of input x. The efficiency measure hj of DMU
measured under this model is the ratio of the total imputed value of its output
levels to the total imputed value of its input level. The total imputed input value is
normalized to some arbitrary level (i.e 1) (Thanassoulis, 1997).
If we consider the multiplier form as the Primal DEA Model, then we can use the
duality in linear programming to compute the equivalent envelopment form of this
problem.
minθ,λ
θ,
s.t Y λ ≥ yi
θxi −Xλ ≥ 0
λ ≥ 0
Where θ is a scalar and λ is a I × 1 vector of constants. This envelopment form
involves fewer constraints than the multiplier form (N + M < I + 1), and hence is
generally the preferred form to solve (Coelli et al., 2005).
The efficiency of a DMU is measured by θ with reference to a Production Possibility
Set (PPS) which contains all correspondences of input and output levels feasible in
the context of the technology operated by the DMU. The PPS is constructed from
observed input-output correspondences at DMUs using a set of postulates (Banker
et al., 1984), hence the inner boundary of this set is a piece wise linear isoquant.
The efficiency measure gives a nice intuitive interpretation, as it measures in effect
the minimum proportion of the observed input levels of the i-th firm which would
be sufficient, under efficient operation, to secure at least its observed output levels.
The practical value of duality in DEA is dependent in the context of its application.
49
The primal model has less intuitive appeal in a value free production context such
as the Superannuation Industry, where its output vectors may not have quantifiable
monetary value. The envelopment form of DEA is therefore more appropriate as
it reflects the extent to which the company can lower its operating expenditure,
given the amount of output it delivers, the number of properties it serves and so on
(Thanassoulis, 1997).
Variable Return to Scale (VRS)
Previous descriptions of DEA linear programming assumes Constant Return to
Scale, earlier discussion in this thesis has revealed that technical efficiency measures
under this assumption is confounded by scale efficiencies (SE) when not all firms
are operating at the optimal scale. Therefore, we need to adapt the existing DEA
model to account for the scale effects. This is accomplished through the addition of
the convexity constraint: I1′λ = 1.
The modified system of linear programming equations is then:
minθ,λ
θ,
st. Y λ ≥ yi
θxi −Xλ ≥ 0
I1′λ = 1
λ ≥ 0,
Where I1 is an I × 1 vector of ones, this constraint essentially ensures that an
inefficient firm is only benchmarked against firms of a similar size. This approach
forms a convex hull of intersecting facets that envelope the data points more tightly
than the CRS conical hull and thus provides technical efficiency scores that are
greater than or equal to those obtained using the CRS model.
50
Scale inefficiencies can be easily isolated using the previously devised formula.
SE (x, q) =di (x, q | V RS)
di (x, q | CRS)=TECRSTEV RS
51
Chapter 5
Empirical Specifications
The specification of output and input variables in production efficiency analysis are
of the utmost importance. This is especially true in DEA analyses, where efficiency
estimates are highly sensitive to the definition and measurement of these variables.
Economic theory suggests that these variables should reflect the entity’s production
process and its key functionalities Ferrier and Lovell (1990).
Two popular conceptualizations of the production process in financial services in-
dustries such as banking, managed funds, and pension funds are: the intermediation
approach and the production approach. The intermediation approach views firms
in these industries as mediators between two parties, their purpose can be seen as
those of an agent that intermediate between investors and prospective investment
projects Humphrey (1985). While the central belief surrounding the production
approach is that firms are services providers, they are seen as fee taking institutions
that provide specialised financial services to its members Kolari and Zardkoohi
(1987). Empirically, the conceptualization of the production process should guide
our definition of output and input factors. The question is then, which production
process is deemed to be more appropriate in the context of this study?
The underlying belief of this study is that superannuation is a crucial part
of Australia’s social infrastructure. Its purpose and objective goes beyond the
intermediation between retirement assets and potential investment opportunities.
Instead, the creation and preservation of asset value should be core to the Industry’s
functionalities. In this sense, the production approach is more appropriate, as it
aligns with the notion that the superannuation industry is a specialised financial
service provider (retirement asset investment) using a variety of input factors (fees
52
and commissions). This approach is also consistent with international pension fund
studies (Basso and Funari, 2003, 2001; Barrientos and Boussofiane, 2005; Barros and
Garcia, 2006).
5.1 Output Specification
Appropriate measures of output factors under the production approach should
closely reflect the key services that the industry provides to its members. Although
the empirical specifications of output variables in the context of pension and
superannuation funds are typically constrained by the availability of data and
adapted to the particular market environment in which the industry operate. Most
studies recognize that the industry performs three key duties: Administration of
accounts, Investments of asset and preservation of value. In the Australian context,
these functionality are in line with the statutory duties of super fund trustees,
therefore they are appropriate for the purposes of this research.
5.1.1 Administration
Administration of accounts typically involves addressing member inquiries, record
keeping, transaction services, reporting duties, and compliance to regulators.
The majority of these functions are incorporated into superannuation licensing
agreements, however additional services such as financial advices, insurance add-
ons and consulting services are also available from some superannuation funds.
The APRA superannuation fund data does not report any information on these
services or their related pricing information. In previous research this deficiency has
been addressed using proxy variables. Mitchell and Andrews (1981) argued that
total number of members could reflect the aggregate service function provided by
the institution since some essential services provided are identical to all accounts.
This was supported by similar studies in the Australian context, where Bateman
and Mitchell (2001) argued the technique is applicable in Australia since the
superannuation market provides near identical products. Indeed, the total number
of members could serve as proxy to the essential account services provided by
53
superannuation funds. However the technique does not reward funds for providing
additional account services that are not common across the industry and is incapable
of capturing the quality of the essential services provided. This limits one ability to
draw inferences on the efficiency scores rankings, since funds that spend additional
resources to provide premium services will be unjustly punished because those high
quality services are not fully captured in the output specifications. The impact of
these limitations is dependent on the extent of heterogeneity of services provided by
the participants of the industry, since these characteristics are not directly observable
in our database, we have decided not to include this output measure in this research.
5.1.2 Investment
Investment services provided by superannuation funds includes portfolio asset
allocations, asset class selections, investment evaluations and executing trades.
To evaluate the fund’s performance in delivering these services, one can take the
perspective of fund members, where the accumulation and creation of wealth is
of the utmost importance. Prior investment performance studies in the broader
financial services literature often adopt index measures such as rate of return, Sharpe
ratio and Jensen’s alpha. Although these measure provide valuable insights into the
relatively performance of the fund, it is not appropriate for the purposes of this
research. In particular, the index measure neglect any scale factors of production,
essentially the model assumes constant return to scale. This restriction limits one’s
ability to isolate the scale efficiency component from technical efficiencies of the
super funds. In this research we shall measure investment performance using net
investment income, which have simple intuitive interpretation; it is the net amount
of wealth the fund has generated for its members in addition to their existing assets.
Net investment income within the APRA superannuation database comprises of two
main components, direct investment income and unrealised capital gains. These
two measures provide a sufficient description of a superannuation fund’s investment
output.
54
5.1.3 Risk Management
Risk management is a common feature for most investment service providers; funds
often make investment choices based on a trade-off between the returns and the
amount of risk involved. Management of risk provides certainty and predictability
to members; assist them in planning and formulating retirement options. Existing
literature has measured risk in terms of variability in returns and also its correlation
with volatilities in the market. This study argues that this is inadequate in
captivating the key features of superannuation funds in Australia. In particular, the
statutory holding period require that distinguishes superannuation funds and other
investment vehicles. As an integral part of our social infrastructure, superannuation
funds have a significant social responsibility in preserving individual’s wealth for
a prolonged period of time until their retirement. A person in Australia can not
access their superannuation saving under age 60. This translates to a fixed long
term investment horizon as opposed to the more flexible investment periods in other
financial service sectors. A comparison of return volatility is somewhat inadequate
in capturing the fund’s performances in reference to its membership characteristics
and their respective investment horizons.
The core objective here is to capture risk exposure of funds in conjunction with
the value of the investment assets and the duration of the risk exposure. One
plausible approach is the Value at Risk measure. The metric incorporate the three
key investment characteristics: expected return, volatility and holding period to
produce a meaningful value measure of losses due to normal market movements.
The application of VaR in other financial institutions has gained tremendous impetus
in growth since the 1990s. Its intuitive interpretation and computational ease are
favoured by private institutions and regulators alike. The application of VaR in
relation to the pension market is especially useful as it measure the potential loss
in portfolio value of members at retirement. Further, age distribution records of
members in the APRA data base allows us to approximate the expected holding
55
periods of assets.
Empirically, this thesis follows the long term Value at Risk framework outlined in
Dowd et al. (2004). Formally the model can be expressed as follows:
V aR (t) = P − exp(µt+ φclσ
√t+ ln (P )
)
Where:
P is the portfolio’s present value.
µ,σ is the mean and standard deviation of the funds’ historical returns.
t is the holding period or investment horizon faces by members.
φcl is the index value of the normal distribution at given probability thresholds.
Figure 5.1 illustrate the VaR measurement of two simulated portfolios at 5%
significance. Subfigure (a) shows a $1 portfolio with low return and high risk, its
terminal value with holding period of 40 years is close to $1. This result suggest
that long term investment under such portfolio is likely to result in the portfolio
value reducing to 0. In contrast, subfigure (b) demonstrate that a similar portfolio
under high return and low risk is likely to have a negative terminal VaR value. This
translates to an overall profit.
As discussed earlier, DEA requires positive and monotonic output measure. To
accommodate this requirement, we perform the following transformation.
V aR′ = (−1)× V aR + 1
56
Figure 5.1: Value at Risk Comparisons
(a) Low Return (4%) and High Risk (0.35)
(b) High Return (10%) and Low Risk (0.15)
The measurement of holding period is especially troublesome, as APRA report age
distribution in discrete intervals. This makes determining the accurate holding
period difficult. However if one assume that members are distributed normally
within the age interval, one can approximate the investment duration using the
median age in the group. A possible criticism of this technique is the violation
of normality in age distributions. For example, it is plausible that in the lower
spectrum of the age categories, age distribution is likely to be skewed to the left, since
membership of superannuation fund is directly linked to work force participation,
57
therefore we would expected a higher density of older workers in the lower spectrum
of age categories.
Another consideration in computing VaR is the measure of return volatility. Given
the granularity of our data (annual observation) and the limited time series (9 years
or less) any standard deviation estimate is likely to be biased. To account for
this caveat in our data, we adopt a similar approach to Ellis et al. (2008), where we
approximate the volatility in return of super funds using a benchmark portfolio. The
benchmark portfolios are constructed by combining the asset allocation information
in our data and their respective market portfolio1. Although this measure is not
accurately reflect the portfolio risk of the fund, we argue it is representative.
Large super fund typically holds extremely diversified portfolio within each asset
categories, asset allocation information are sufficient in determining the return
volatility of such super funds. The caveat of this technique lies with smaller funds
where the market portfolio is not a good proxy for risk.
It must also be noted that some superannuation funds outsource their investment
services to an administrator. It can be argued under these circumstances funds are
no longer producing the entirety of its outputs. Therefore, any analysis of fund
efficiency will also endogenously incorporate those of its administrator. Here we
argue that even in such scenarios,a fund does not simply proliferate itself from its
duties; rather, this represents a shift from ’in house’ to a ’proxy’ production process,
where funds are responsible for the selection of investment agents (administrators)
as well as instructing their investment philosophies. Efficiency analysis of funds are
thus still valid despite a slight change in its intuitive interpretation.
5.2 Input Specification
Similar to that of the output production factor definitions, input factors are selected
based on the hypothesized production process of superannuation funds, in which
1See appendix for a table of volatility measures for the market portfolios
58
input factors should be directly utilised for the provision of services to members.
This paper has defined the input factors using total assets under management and
total expenses.
5.2.1 Total Assets
In line with the production approach, firms requires the input of capital and labour.
In our institutional setting, capital input could be captured using the total value of
assets under a fund’s management. Funds invest existing assets in a variety of classes
including fixed incomes, equities, properties and other assets to generate investment
income. To some degree, the variable is also capable of capturing the scale effect
of market access; for example, funds with a greater asset pool is believed to have
access to a greater spectrum of investment options than smaller funds. One would
expect that this effect is insignificant once a certain threshold of asset holding is
achieved.
5.2.2 Administrative Expenses
Labour inputs are typically measured by wage expenses in traditional economic
literature. In the financial services sector, these most closely translate to account
administrative expenses, and investment related expenses. These figures are directly
available from the APRA’s fund-level superannuation database, however one should
be cautiously aware of how these expenses are incurred and reported to the regulator.
Administrative expenses are typically incurred from staffing of ’in house’ service
functions such as record keeping, regulation compliance, transactions and other day
to day operations. The reported administrative expenses to APRA are believed to
be fairly accurate. A potential source of complication arise in relation to marketing
costs; for example, non-public offer funds such as those in the public sector and
corporate fund require relatively low marketing activity, while public offer funds in
the retail sector need to spend substantially more on advertising and other publicity
59
campaigns. The difference stems from a range of factors including fund type,
corporate governance and other circumstances that are not under direct control of
the fund. Since the available data does not allow us to isolate marketing expenses,
this represents a potential sources of endogeneity. where funds that require intensive
marketing activity are punished for something for which they have no direct control.
Robustness checks will be performed on efficiency scores to assess the sensitivity of
our results to these issues.
Investment related expenses accounts for a broader category of costs associated
with the investment process. These costs could arise from labour costs in
strategy formulation, evaluation of securities, consultancy as well as transaction
costs. However due to outsourcing, the reported investment expenses in the
APRA data is highly unreliable (Coleman et al., 2003).For example, when funds
outsource investment activities to a third party, they often report the net investment
return to the regulator and report nil investment expenses. This presents similar
interpretation issues due to outsourcing to those from the previous discussion on
the output definition. In addition, the reported nil investment expenses presents
statistical challenges to the linear programming of DEA, where efficiency of these
firms are likely to be skewed upwards due to the program’s inability to find
comparable peers. One plausible solution is to use total expenses, where we aggregate
investment and administrative expenses. The measure ensures that the input factors
are strictly positive and provides a simple intuitive interpretation. However, this
aggregation does limit one’s ability to isolate the specific slacks in input factors
which restrict inferences from results.
60
Chapter 6
Data
The primary source of data used in this study is the annual superannuation fund
level publication released by the Australia Prudential Regulatory Authority (APRA)
for the financial years 2004 to 2012 inclusive. This data is compiled from annual
surveys for regulatory purposes and is publicly available on the APRA’s website.
Funds under APRA’s oversight recorded in the database represent a substantial
portion of the current retirement assets under management, with the exclusion of
small Self Managed Superannuation Funds (SMSF), Pooled Superannuation Trusts,
Approved Deposit Funds and Exempted Public Sector Funds. The sample remains
highly representative of the industry, with more that 59.5% of assets captured in
2012 and higher percentages in earlier periods. This section will discuss the issues
related to the data, especially the sample selection process and the limitations in
the measurement accuracy.
6.1 Variables
The reported information in the data set can be broadly divided into two categories:
financials and characteristic data. This conveniently aligns with the objective of
this research, where financial information is used to construct efficiency scores; we
then investigate the impact of a variety of fund characteristics on the operating
efficiencies.
6.1.1 Financials Information
There is a substantial amount of financial information in the data set. However for
the purposes of this study, variables relating to investment returns, expenses and
asset holdings are of particular interest.
61
Total Assets
Assets is mainly used as the portfolio value in Value at Risk calculated, the desired
variable is asset holding by account type, which was categorized by gender as well
as age groups. The age group intervals are: (1) Less than 35 years old, (2) Between
35 to 49 years old, (3) Between 50 to 59 years old , (4) Between 60 to 65 years
old and (5) Above 65 years old. These decompositions of asset holding allows us to
compute more accurate Value at Risk measures as the vested exposure is specific to
the particular age group.
A particular issue associated with the asset variable is the reconciliation between
vested assets and the net asset measure. In some instances, the total vested asset
is lower than the new asset reported; this is likely due to legacy Defined Benefit
(DB) asset holdings, which is not accounted for in the vested asset measure. In
some cases, this difference is substantial, and could bias the results as the assets
under exposure is understated. To account for this difference, we assume that the
DB asset is distributed evenly across all age groups; this allows us to simply rescale
the vested asset figures for each age group using their current proportion of asset.
The validity of the uniform distribution assumption is weak, however due to the
lack of further information and its limited implications, we would argue that the
assumption is innocuous.
Investment Outcomes
Net Investment Income is another key variable that is central to the efficiency
analysis. This variable is directly available from the APRA data set and it is
typically composed of direct investment income and unrealised capital gains. Direct
investment income includes dividends, rents, trust distributions and interests. Bad
debt expenses are also deducted from this measure. Note that the proceeds of
insurance policies proceed does not contribute to this figure.
62
Fees
For reasons outlined in the Empirical Specification Section, Total Expenditure is
used instead of its individual components of administrative and investment expenses.
6.1.2 Structural Characteristics
The data set captures several structural dimensions of the superannuation industry
in Australia. This include fund type, benefit structure, membership and public offer
status.
Fund Type
Fund type in the superannuation industry typically reflects its governance structure
as well as the sector in which it operates in. There are typically 5 main fund
types: (1) Corporate Funds, (2) Industry Funds, (3) Public Sector Funds, (4)
Retail Funds and (5) Others. Also note that first three categories of super fund
are also ’not for profit’, which means that they charge members for services at
cost. Of the funds categorised as ’others’, majority belongs to lost and inactive
superannuation accounts, these accounts are unclaimed and are set aside by funds
in passive management; due to it’s relative menial market share and the nature of
the accounts, funds categorised as ’others’ have been excluded from this study.
Benefit Structure
Benefit structure is another categorical variable that includes: (1) Defined Contri-
bution (DC) Funds, (2) Defined Benefit (DB) Funds and (3) Hybrid Funds. Hybrid
funds are typically legacy DB funds that have closed their DB options to new
members and offers only DC options. For the purposes of this study, hybrid funds
are relabelled as DC funds. This is in line with their close resemblance to that of
the DC funds and their diminishing functions as DB fund provider.
63
Membership
The data offers several membership related information with variables which includes
age and gender distributions of accounts; and the concentration of assets in the
default investment strategy.
Age and gender distribution of accounts is reported in identical format and intervals
as that of the asset distribution.
Asset concentration in default measures the percentage of asset in the super fund’s
default strategy. This measure could be used to proxy the level of member
engagement in the funds. This is in no way a perfect proxy, as default strategy
selection could be attributed to both inactive and active selection. Where members
elect the default option because it’s the best option can’t be isolated from those who
fail to chose investment options. Further, funds without default options report the
investment option with the highest assets as their default strategy, so the measure
is misleading in such circumstances.
Other variables also include: number of active members and the number of
investment choices available.
Public Offer Status
Public offer is a binary variable that describes whether a fund is open to members
of the general public. For example, public sector and corporate funds are typically
non public offer funds, since members who are not part of the Australia public
service or the corporation in which the fund belongs to will not be able to join these
funds. Unlike other structure characteristics of the funds, public offer status is less
persistent; recent trend suggest that many funds are opening up to the public.
64
6.2 Sample Selection
Due to DEA’s sensitivity to the sample being analysed, a vigorous sample selection
process is conducted. where unreliable and erroneous observations are dropped from
the sample.
6.2.1 Erroneous Data
Erroneous data are observations that are empty or those that does not make logical
sense. These observations serves little purpose and may potentially bias the results,
therefore they are dropped from the analysis.
Where key variables including, investment return, expenses and assets are missing
the observations are dropped.
Observations with negative expenses and assets are also dropped from the sample
as these measurement are not possible.
In 2004, APRA made changes the reporting unit of some key measures, including
total assets, contributions etc (from $ to $’000). Correction are made for selected
observations in the 2005 sample which did not take into account of this change.
6.3 Final data
Table 6.1 provides the summary statistics of the final data set used in this study.
65
Table 6.1: Summary Statistics
Category Observations Units Mean Std. Dev Min Max
Investment Income Output 2807 $ m 76.71 453.71 -5694.29 5351.33
Value at Risk Output 2807 $ m -9484.06 40700.00 -947000.00 26000.00
Total Expenses Input 2807 $ m 13.96 34.69 0.00 370.41
Total Assets Input 2807 $ m 1887.62 4949.27 0.06 51900.00
Investment Choices Enviromental 2807 choices 55.10 239.13 1.00 2829.00
% Asset in Default Enviromental 2807 % 56.65 36.66 0.00 100.00
Value at Risk are raw, pre transformation measures.
66
Chapter 7
Results
The efficiency analysis in this research can be broadly categorised into two stages.
First, a set of technical efficiency scores are constructed using DEA and initial
observations and comparative studies are conducted on the estimated efficiencies.
The second stage analysis aims to examine key drivers of industry inefficiency
under the regression framework, where several popular second stage models from
the literature are assessed to ensure consistency and robustness of our results.
7.1 Limitation of Inference
Prior to any analysis, it should be noted that any statistical inference based on
DEA analysis should only be interpreted subject to the underlying assumptions.
In this research, we analyse superannuation funds in Australia using two output
factors: Investment Earnings and Value at Risk, and two related input factors:
Administrative Expenses and Total Assets. This specification reflects the quantifi-
able financial performance of super funds from the member’s perspective. However
this interpretation does not necessarily resemble the true performance of the super
funds as there are other dimensions of success that are incorporated in this model,
including customer satisfaction, the diversity and quality of other services offered, as
well as the other unique core objectives of the entities in the market. Therefore our
results of super funds’ efficiency performance are limited to our particular empirical
specification.
Further, all outputs and inputs are normalised relative to the mean for a given
period t. In the context of DEA analyses, this manipulation rebalances the scale
and hence the relative weighting assigned to funds’ output variables and input
variable are identical. However, in reality this does not necessarily reflect the true
weighting of various factors in a super funds’ production process (that is, the relative
67
importance of factor inputs/outputs). Unfortunately due to the atypical nature of
the production process and the granularity of our data, the true production factor
weight is difficult to determine. Given these limitations, it is reasonable to assume
equal production factor weight, so inferences based on efficiency should also take
this under consideration.
A reminder is also necessary in relation to the interpretation of results. This thesis
took an input orientated DEA approach, therefore an efficiency score of say 0.80
can be interpreted as meaning that this super fund could reduce weighted input
factors by 25%(
1−0.80.8
= 0.25)
without altering output levels. Also recall that overall
technical efficiency (OTE) is a product of its constituent components: pure technical
efficiency (PTE) and scale efficiency (SE). The decomposition allows insights into
the sources of inefficiencies. It also helps determine whether super funds have been
operating at their most productive scale size(MPSS), increasing returns to scale
(IRS) or decreasing returns to scale (DRS).
7.2 First Stage: Efficiency Scores Analysis
The linear programming program described in previous chapters is applied to the
data and the technique is iterated for each year to account for the movements
in the frontier and market conditions. From this, a series of efficiency scores are
generated for each super fund. In presenting these results, we first turn to the
industry averages, as this measure is a good reflection of the overall performance of
the super funds in the market.
68
Table 7.1: Average Efficiency Scores
OTE PTE SE Funds
Year Average St. dev Average St. dev Average St. dev
2004 0.82 0.06 0.94 0.07 0.88 0.06 472
2005 0.84 0.05 0.93 0.07 0.91 0.05 425
2006 0.83 0.07 0.93 0.08 0.90 0.07 336
2007 0.84 0.05 0.94 0.06 0.89 0.06 322
2008 0.77 0.26 0.78 0.26 0.98 0.06 304
2009 0.76 0.26 0.77 0.26 0.97 0.06 280
2010 0.88 0.09 0.92 0.10 0.95 0.04 251
2011 0.90 0.12 0.95 0.07 0.95 0.08 218
2012 0.79 0.24 0.83 0.21 0.94 0.13 199
Perhaps the most striking result from Table 7.11 is that industry average OTE
has been relatively high throughout the period, but there is a noticeable decrease
during the GFC period between 2008 and 2009. This appears to be primarily driven
by a substantial decrease in the PTE; the opposite movement in scale efficiency
has somewhat reduced the impact of PTE movements on overall technical efficiency.
Further, downward movements in the PTE average is coupled with greater dispersion
of scores across the industry. A plausible explanation for these observations is related
to the Global Financial Crisis (2008-2009), where the majority of superannuation
funds in the Australian market reported significant investment losses, whilst their
input vector such as fees remains relatively unchanged. In these circumstances, the
super funds that did generate positive returns or slightly less portfolio loss during this
period will appear significantly more efficient than the rest of the industry, which
results in larger dispersion of efficiency scores towards the lower spectrum. This
chapter will focus on the movements in the PTE and the impact of environmental
variables including super fund governance structure, regulatory framework, benefit
design and investment choices.
1OTE : Overall Technical Efficiency (CRS efficiency); PTE : Pure Technical Efficiency (VRSefficiency); SE : Scale Efficiency.
69
The survey of the Australian literature revealed only one study that had previously
measured technical efficiency of superannuation funds during this period. In contrast
to our results, Sathye (2011)’s analysis of retail funds from 2005 to 2009 found that
the decrease in average OTE occurred in 2009, with scores averaging 0.16 from 2005
- 2008 and 0.07 in 2009. This is an interesting result and its difference to the results
presented here may partly be explained by the differing specification of inputs and
outputs factors. In his study, Sathye (2011) has specified ’value of the fund’ as
one of two outputs. This variable is likely to be related to our investment income
variable in a one period lag structure, where large investment losses are realised in
the ’fund value’ measure in the subsequent period. This helps explain why Sathye
(2011) has found decreases in efficiency from 2009 instead of 2008 as per our results.
Further, Sathye (2011) has only examined the efficiencies of retail superannuation
funds, whereas we have examined all super fund types except for self managed super
funds (SMSFs). Since DEA analysis is extremely sensitive to reference units in the
sample, this difference is likely the primary source of differences between our results.
Recall that pure technical efficiency (PTE) measures how well the super fund
manages its resources to deliver its vector of outputs. Our results show that
Australian super funds perform relatively well on average, with a 9 year average of
0.89. The higher average does not necessarily imply that Australian super funds
on average are highly efficient. However this clustering around the production
frontier does show that on average Australian super funds perform very closely
to the best performing entity in the market. Given the unique production process of
the superannuation sector and the lack of comparable benchmarks, it is reasonable
to use the best performing units in the market as the reference points to infer
results. Regardless, under input orientation DEA, our result translates to a
potential reduction of 12.2% in input factors if the industry operates at its maximum
production efficiency. Further, since super fund are limited in manipulating their
input factor ’Total assets’, potential input oriented efficiency gain is most likely
70
achieved from cost reductions (i.e decreasing input ’administrative costs’). The
exact quantifiable benefit is dependent on the unique input slacks of each super
funds analysed in our sample. This will be explored in a case study context later in
this chapter.
Our results also show that there are limited scale efficiency gains; with a 9 year
average of 0.93, this suggests that on average, Australian super funds appears to
operate very close to their most productive scale size. Despite this, our results
show that in 2012, 59.3% of Australian super funds are operating at increasing scale
to return, that is these funds could somewhat improve their efficiency by increasing
their scale. It is comforting to see that the majority of these super funds are industry
funds(35.59%) and retail funds(34.75%), as other fund types are somewhat limited
in growth due to their restrictive membership base. Our previous discussion showed
that an increasing number of corporate funds are merging into retail trust funds or
industry funds, and under these circumstances we would expect the super industry
to somewhat benefit from scale efficiency improvements in industry and retail funds.
7.2.1 Efficiency Rankings
Efficiency analysis in its essence is a form of benchmarking technique, and allows
us to identify how super funds perform relative to one another. As such, it is
natural that we rank industry participants using their efficiency performance.In
this research, we have defined industry leaders as those funds that systematically
obtained top scores in terms of their productive pure technical efficiencies. The
results are presented in Table 7.2. An initial review of market leaders did not reveal
any prevalence by a particular superannuation fund type; that is, there appears to
be a mixture of Corporate, Industry, Public Sector and Retail funds.
71
Fund Name Average Efficiency Type Frontier
Betros Bros Super 0.9992 Corporate 5
Commonwealth Life Personal 0.9328 Retail 5
MLC Superannuation 0.7793 Retail 5
Australia Post Super 0.8644 Public Sector 4
UniSuper 0.7714 Industry 4
BT Superannuation 0.9978 Retail 3
Construction & Building Union 0.7561 Industry 3
Perhaps the more striking result Table 7.2 reveals is that all three retail funds
that frequently appeared on the production frontier are staff superannuation funds
associated with the large commercial banks in Australia - that is, Commonwealth
Life Personal, MLC and BT super are affiliated with the Commonwealth Bank
(CBA) National Australia Bank (NAB) and Westpac respectively. This result is
consistent with our a priori that large financial service entities have the ability to
leverage their expertise and overheads from their core financial services activities
into the superannuation management service of their own staff super funds. In
essence, this translate to a lower operating cost base, making these funds highly
competitive and very efficient.
Our results also show that while some funds have frequently appeared on the frontier,
they have a relatively lower average efficiency score than similar frontier super funds.
This result can be primarily attributed to the poor performance during the 2008 -
2009 GFC period, when these funds significantly under-performed their benchmark
reference units.
As an example, we conduct the following case study. UniSuper, the superannuation
fund for employees of the tertiary education sector, is one of the largest industry
72
funds in Australia; in 2012, UniSuper managed over 470,000 accounts and approx-
imately A$32.5 billion in assets. This gives UniSuper 4.13% of the total market
share in Australia’s superannuation sector by assets. Our DEA analysis shows that
UniSuper has consistently performed well in 7 year period, excluding 2008 and
2009, appearing on the production frontier in 4 out 7 years and scoring an average
efficiency of 0.96. However, UniSuper’s efficiency significantly decreased during the
GFC period, averaging only 0.11. This can be primarily attributed to its large
investment losses during this period (-AU$1.50 billion in 2008 and -AU$2.13 billion
in 2009). A set of reference funds around the target fund from which the hypothetical
efficient unit on the frontier is constructed is provided by the DEA program for each
year. Table 7.3 shows the reference funds for years 2008 and 2009.
Table 7.2: Reference Decision Making Units for Unisuper
2008 Reference Units 2009 Reference Units
Towers Watson Towers Watson
BMA Personal Tower Australia
Standard Communications
A common feature among the funds in the frontier reference set during this period
is that either they have managed to generate positive investment income during the
GFC period, or recorded relative small losses with low expenses. As an example,
BMA Personal recorded a 5.7% net return and an expense ratio of 0.7% in 2008;
whilst UniSuper recorded a 6.2% net loss and an expense ratio of 0.5%. Similar
comparisons can also be made with the second reference point (Towers Watson).
From these figures it is clear that a notable source of inefficiency is from investment
income outputs. Indeed, DEA’s administrative expenses input slack estimate for
UniSuper is around 19.5%; that is, if UniSuper employed strategies or performed
similarly to the two reference funds, the fund could reduce its operating cost in this
73
period from 0.7% to approximately 0.56%.
Further, our DEA analysis also shows large inefficiencies derived from the Risk
Management Output (VaR). This is could be attributed to both investment return
and differences in portfolio volatility between UniSuper and the reference funds.
In 2008, the super fund BMA Personal had 100% of its asset holdings in cash;
while the Towers Watson super fund held 53.6% of its assets in domestic fixed
interest securities and cash.2 In comparison, UniSuper has substantially higher
exposure in the equities market and relatively low holdings in bonds. These portfolio
allocation differences combined with the volatile financial climate during the GFC
period contributed greatly to the substantially higher portfolio value at risk figure
estimated for UniSuper in comparison to the reference super funds.
Another interesting feature of super funds in the reference set is the limited number
of investment options offered. Both funds in the 2008 reference set only offered
one investment option to their members; this correspond to extremely high levels of
asset concentration in the default strategy (BMA (100%), Towers Watson (93%)).
Since our a prior information suggests that fund managers have significantly greater
control over their default strategy, it can be argued that this characteristic allowed
these funds to more easily manage their various asset exposures and more effectively
adjust to any unexpected market movements. In the case of the Towers Watson super
fund in 2008, the fund had adjusted its combined domestic fixed interest security and
cash holding up from 38.3% in 2007 to 53.6% in 2008, international equity holdings
have also decreased by 3.2%. These asset allocations movements are not uncommon
during this period. However the high level of portfolio management concentration
grants fund managers with greater flexibility and allows any changes to propagate
through all the member accounts at a faster rate. In contrast, UniSuper is fairly
decentralised in their portfolio management, with 9 available investment options
2Asset holdings categorised as ’others’ are assumed to be some kind of cash holdings, this isconsistent with assumptions made in Ellis et al. (2008).
74
and only 26.1% of assets in its default strategy. Similar to the two reference funds,
Unisuper has also increased its fixed income and cash holding during the GFC
period, however it’s reaction was much more subtle, with a modest 2.8% increase
from 30.4% in 2007 to 33.2% in 2008 (compared to 15.3% increase in the Towers
Watson super fund). This is consistent with our a priori that decentralised super
funds’ reaction to market movements are dependent on the active management of
members to some extent. While these super funds are able to offer more conservative
investment options during market downturns, it is ultimately up to the members
to divert their portfolios into these investment strategies. Given the low level of
membership engagement evident in the Australian superannuation enviroment, this
feature of the portfolio management chain is likely to result in lagged response of
some super funds to shocks in the financial markets.
While this evidence seems to support our previous assertion that greater portfolio
concentration enables super funds to more effectively reaction to market shocks
during periods of financial distress, the implication of this is less clear in relatively
modest periods. The issue here is then the trade-off between super funds’
ability to effectively manage assets, and the utility different members derive
from their portfolios. Increasing Academic literature analysing the Australia
superannuation market has concluded there is a need for more tailored investment
choices, accommodating for individual’s risk preference structures, age, and other
characteristics. Greater investment choice could facilitate such demand, however
this is likely to decreases the fund manager’s ability to effectively control the asset
composition of the fund itself. More sophisticated analysis is required to determine
the optimum level of portfolio choices, however this is out of the scope of this
research.
The identification of efficiency leaders could potentially be very useful to market
participants and regulators alike. Careful study of consistently efficient firms could
75
provide valuable guidelines for further regulatory reform, informing policy decisions
and to promote ”yardstick” competition between market participants. Moreover,
previous industry studies have recognised the importance of a single performance
measure in reducing the complexity of the information presented to consumers
through its role in promoting consumer engagement and market competition.3
Efficiency scores are an intuitive method to facilitate such discussion; the technique
is also advantageous as its input/output vectors could be altered to accommodate
different research objectives. While this research has focused extensively on the
efficiency of super funds in delivering benefit for members, one could easily alter the
output variables to examine how efficient the superannuation industry is operating
in alleviating public pension liabilities.
7.2.2 Public Offering
The analysis thus far has considered average efficiency across the industry as a
whole, however this research also has particular interest in examining the differences
in efficiency performance across sectors and fund types. Table 7.4 summarises the
efficiency trends between public offer funds and non public offer funds. Recall that
funds on public offer are available to the general public while non public offer funds
are closed funds are only available to employees of a particular firm or industry.4
3Coleman et al. (2003)4The majority of public offer funds are retail funds while closed funds are mainly comprised of
industry, corporate and public sector funds.
76
Table 7.3: Average Pure Technical Efficiency: Public Offering
Non Public Offer Super Funds Public Offer Super Funds
Year Efficiency (PTE) Scale Efficiency Efficiency (PTE) Scale Efficiency
2004 0.96 0.87 0.91 0.90
(0.05) (0.06) (0.10) (0.07)
2005 0.95 0.89 0.91 0.92
(0.05) (0.05) (0.08) (0.06)
2006 0.96 0.88 0.90 0.91
(0.04) (0.06) (0.09) (0.07)
2007 0.95 0.88 0.93 0.90
(0.05) (0.05) (0.06) (0.06)
2008 0.86 0.99 0.71 0.97
(0.18) (0.05) (0.29) (0.07)
2009 0.86 0.98 0.71 0.97
(0.18) (0.04) (0.28) (0.07)
2010 0.95 0.95 0.91 0.95
(0.07) (0.03) (0.11) (0.04)
2011 0.97 0.97 0.93 0.94
(0.05) (0.06) (0.08) (0.08)
2012 0.93 0.96 0.79 0.93
(0.09) (0.12) (0.24) (0.13)
Standard deviations in parentheses
On average, closed super funds appear to be more technically efficient than funds
on public offer. This is in line with our prior, where public offer funds are believed
to be less productive than their counterpart due to additional expenses arising
from marketing, distribution and other related costs. A more striking result is
the relatively low scale efficiency of the non-public offer (closed) super funds. These
super funds by their nature are constrained to the members in their constituent
industry or corporation. For example, an industry fund such as UniSuper is only
available to university employees. This underlying structural feature could prevent
super funds from reaching their maximum productive scale size, hence operate at a
lower scale efficiency. Our results show that average scale efficiency of non-public
offer funds converge with their industry counterpart over time. This result is also
consistent with the observation that an increasing number of industry funds are
77
becoming public offer funds in order to expand their membership base. Further
analysis reveals that the majority of these industry funds previously operated at
increasing scale of return, which suggests offering their membership to the general
public is likely to improve their overall efficiency. These results are somewhat
supported by the relevant literature; Bateman (2001) showed that public offer retail
funds exhibit the highest level of administrative fees.
7.2.3 Fund Types
This research is also concerned with the efficiency performance of the four fund types
currently available in the market - that is corporate, public sector, industry and retail
super funds. Our results are presented in Figure 7.1, which shows that corporate
funds consistently outperform the market in pure technical efficiency (PTE).
Figure 7.1: Average Pure Technical Efficiency: Fund Type
.6
.7
.8
.9
1
Ave
rage
Effi
cien
cy S
core
2004 2006 2008 2010 2012Year
Industry CorporatePublic Sector Retail
Perhaps the more interesting result from this exercise is the dynamic of technical
efficiency movement during the financial crisis (2008 - 2009). Here we observe
a substantial decrease in average efficiency across the industry, with the effect
much more severe among public sector and industry funds. Again corporate funds
78
outperformed other funds types, followed by retail funds. This result somewhat
reflects the ability of funds to manage market externalities, in particular shocks
in the financial markets. Higher efficiency during this period can be generally
attributed to good performance in delivering outputs (Investment Income and
Portfolio exposure) and/or the reduction in inputs (Administrative costs). Further
analysis revealed that in general, more efficient super funds during this period
typically have relatively higher holdings in fixed income assets and cash. By passively
investing in these securities, some super funds were able to limit their portfolio losses
while maintaining relatively low levels of administrative costs. Also, one should keep
in mind that super funds such as industry and corporate funds generally have a
greater proportion of members in their respective default portfolios, where default
portfolios in Australia are typically balanced investment options, with 30% invested
in bonds and 70% in shares. In contrast, retail super funds typically experience
a relatively higher degree of member engagement and prior to the introduction
of MySuper5 were less likely to offer default options, whereby members are more
actively involved in the asset allocation process. Higher level of membership
engagement coupled with the flexibility offered by decentralised portfolios could
explain why retail funds are more efficient during periods of financial crisis.
During the non GFC periods (2004 - 2007, 2010 - 2012), the three remaining sectors
of the industry are generally indistinguishable in their efficiency performance, with
public sectors funds slightly outperforming others in 2006 and 2012.
7.2.4 Efficient Market
Goddard et al. (1993) argued that more efficient firms display superior performance
and are more viable in a competitive environment. In a rational and efficient market
setting, one would expect more efficient super funds to generate higher output and
increase their market share at the expense of less efficient funds, thereby increasing
5MySuper is an industry reform initiated by the Australian federal government in 2010, one ofits key policy is the introduction of compulsory default option.
79
industry concentration. The literature has conceptualised this idea as the ”Efficient
Structure Hypothesis”; an implication is that efficient and inefficient firms cannot
coexist in the long run. The hypothesis that technical inefficiency increases the
probability of firm exits has been well documented in the literature (Kumbhakar
et al., 2009; Tisonas and Papadogonas, 2006; Wheelock and Wilson, 2000). The
causal link between inefficiency and market exit remains an open research question.
We conduct a preliminary testing of the efficient structure hypothesis by computing
the average efficiency scores of super funds that are leaving the market6 and super
funds that remained at least for another year. Our results are presented in Figure
7.2.
Figure 7.2: Average Pure Technical Efficiency: Market Exits
.75
.8
.85
.9
.95
1
Ave
rage
Effi
cien
cy S
core
2004 2006 2008 2010 2012Year
Wind Up Funds Incumbents Funds
Contrary to our priori beliefs, our results show that average pure technical efficiency
of wind up super funds are consistently higher than incumbent funds, except in
2011. This is not a surprising result given the market structure and features of the
Australian superannuation industry. In particular, these results have identified two
6Market exits in the superannuation context means that the super fund cease to exit as aseparate entity, but its assets and members are merged into another existing super fund.
80
key areas where the efficient structure hypothesis may fail.
First, the hypothesis implicitly assumes that all firms within the industry compete
for market share, and that more efficient firms are able to leverage their efficiency to
be more price competitive than less efficient firms, hence expanding output through
acquisition of market share. However, the Australia superannuation environment
presents a different reality. For example, our results have shown corporate funds
to be much more efficient than the rest of the industry. However these super funds
are almost always restricted to the employees of the sponsoring firms. This feature
essentially constrains the potential membership base of the super funds to a small
sub population, significantly limiting its ability to gain any substantial market share
from less efficient incumbent super funds. Further, corporate funds are commonly
not-for-profit. This indicates that the funds operate at cost, therefore there is
limited incentive for the funds to expand its membership base. This conclusion
is consistent for all not for profit, non public offer funds. This is a powerful result
as it indicates that the Australian superannuation industry suffers from structural
limitations where market allocative productivity gains are somewhat limited to
public offer super funds.
The second caveat of the hypothesis arises from cost subsidization. Selected super
funds in Australia (mostly corporate) are heavily subsidized by their respective
sponsors. This then gives rise to the question whether our productive efficiency
measure truly reflects the productivity of the super funds’ operations. This question
has significant implications. In particular, as Figure 2.3 illustrates, over the
past decade, the majority of market exits in the industry has been corporate
funds. Previous research has attributed these exits to increasing operational costs
associated with regulatory requirements, and the reduction of employer’s willingness
to subsidize the costs of their corporate superannuation funds.7 If this is the case,
7Cost subsidization can be in the form of monetary donations to the fund, as well as donationof personnel, executive time and resources
81
then it can be argued that some of the efficiency of these super funds can be
attributed to the amount of cost subsidization, and the true productive efficiency
may well lie below the industry average. However, the analysis here is not concerned
with the true productive efficiency of the super funds, but rather their efficiency in
delivering outputs for members. Therefore, we proceed our analysis by assuming
cost subsidization to be part of the super funds production process, where it can be
thought of as a discount in factor prices available to selected number of funds.
7.3 Second Stage: Regression Analysis
Two stage analysis efficiency scores is popular in the existing DEA literature.
Indeed the procedure is very appealing both in terms of its simplicity and the way
efficiency is described and interpreted. We are interested in the effect that exogenous
environmental variables including fund structure, consumer engagement and market
share have on the pure technical efficiency (PTE) of superannuation funds.8
7.3.1 Regressors
The exogenous environmental variables used in this study are a combination of
extracted APRA fund characteristic data and generated variables.
Fund Type
APRA categorise funds into four main types: (1) Industry Funds; (2) Corporate
Funds; (3) Public Sector Funds; (4) Retail Funds. The categorical string variable
in the APRA data is converted to three binary dummies, setting retail funds as the
base dummy.
Regulatory Type
The binary dummy variable describing the regulatory environment in which the
fund operates in: = 1 if fund membership is open to the general public (i.e,
a public offer super fund); The alternative = 0 are funds that have restricted
8The dependent variable PTE is continuous and confined to the unit interval (excluding 0).
82
membership to particular sub populations such as corporate employees or public
servants (i.e, a non-public offer super fund).A key difference between public offer
and non-public offer is the system of governance. Public offer funds are governed by
corporate trustees, where the trustee board comprises employees of the managing
financial institution. A non-public offer fund trustee board has equal representation
of employer and employees representatives (although two can be supplemented by
independent trustees).For example, the UniSuper board of trustees has 4 employee
representatives, 4 employee representatives and 3 independent trustees.
In recent years many not-for-profit industry super funds which have become public
offer funds, but retained equal representation on the trustee board. Therefore
we would expect some degree of heterogeneous effect of this variable on industry
funds, interaction term between public offer and fund types are also included in the
regression framework.
Benefit Type
A binary dummy variable that describes funds’ benefit type; = 1 if the super
fund is a pure Defined Benefit (DB) fund; that is, members accrue retirement
benefits using a prescribed formula involving final salary, years of service and some
multiplier. The base alternative are funds that have pure Defined Contribution (DC)
or Hybrid DB/DC structure, DC members accumulate wealth based on their life
time contribution and funds’ investment performance. Hybrid funds are difficult to
classify, however the majority are legacy DB funds that have closed to new members
and are phasing into a DC structure. Therefore it is reasonable to assume that they
associate more closely with DC funds.
Members in the Default Investment Option
A continuous variable in [0,100] describes the percentage of assets who in the default
portfolio provided by the super fund. Funds that did not nominate a default portfolio
83
reports the percentage of members in their largest portfolio by asset. In essence, this
variable measures the degree of concentration in the fund’s portfolio management.
Market Share
A continuous variable in [0,100] describing the market share of the fund by assets
in a particular year.
Our a priori knowledge suggest that market power could have a heterogeneous effect
across different fund types. Therefore we also include interaction terms between this
variable and the fund type dummies.
Number of Members
A continuous variable describing the logarithm of the number of members in the
fund at a particular year.
Number of Investment Choices
A continuous variable that describes the logarithm of the number of investment
choices offered by the fund in a particular year.
Market Exit
A binary dummy that =1 if the fund no longer exist in the subsequent year. =0 for
all funds in 2012 since this is the last year of our data.
Recall that our previous result have attributed corporate exit to the inability of the
employers to further subsidize their super funds. We then would expect some level
of heterogeneity between corporate fund exits and other market leavers. Interaction
between exit dummy and corporate fund dummy is also included in the regression.
84
Herfindahl Index
A continuous variable in [0,100] describing the level of market concentration in a
given year. The variable is measured as the squared sum of market shares of all
super funds in the market.
7.3.2 Model Selection
Second Stage DEA efficiency score regression is challenging due to the unique
nature of the dependent variable (PTE efficiency scores) being continuous in ]0,1].
Standard linear models are generally not appropriate for such analysis due to their
unboundedness. A popular approach among the DEA empirical literature is the two
limit TOBIT (2LT) model bounded at zero and unity. Recent developments in the
literature saw Fractional Response Models (FRM) rise to popularity, these models
are difficult to implement but are more flexible and accurate in accommodating
the underlying features of the data. The correct modelling approach remains
controversial, and this section of the paper will discuss the merits of three most
popular models employed in the literature.
Before we start, it is useful to note the key selection criteria in assessing available
models. The DEA technique measures efficiency relative to a non-parametric,
maximum likelihood estimate of an unobserved true frontier, conditional on observed
data resulting from an underlying data-generating process (DGP). A trend in
the previous literature was to regress the efficiency estimates on environmental
variables in a two stage framework. This procedures allows researchers to account
for exogenous factors that might affect firms’ performance, however most studies
lack proper consideration to the functional form of the DGP implicitly assumed by
their models. A coherent DGP description is essential to the correct specification
of the regression model in the second stage and to ensure that the estimates are
consistent. Further, statistical inference is complicated by the unknown serial
correlations between estimated efficiencies. This section of this thesis will draw upon
85
the experience of the current two stage DEA literature and debate the appropriate
model for our context.
Let y be the variable of interest (PTE efficiency scores), 0 < y ≤ 1, and let x be
a vector of k environmental factors. Further, let f(y|x, θ) denote the conditional
distribution of y, which is unknown, where θ is a vector of parameters to be
estimated.
Linear Probability Model (LPM)
Typical LPM models are characterised by the conditional mean, in which:
E[y|x] = xθ,
However, these models employ OLS estimators and are often criticised for two
reasons. First the conceptual requirement that the predicted values of y lie in
the interval ]0,1] is not satisfied. Second, in a linear model, the marginal effect
on the DEA score of a unitary change in covariate xj is constant over the entire
range of y, which is not compatible with either the bounded nature of DEA scores
or the existence of a mass point at unity in their distribution (Ramalho et al., 2010).
∂E(y|x)
∂xj= θj,
Despite these shortfalls, OLS could still be implemented if it’s appropriate for the
underlying DGP. McDonald (2009) showed that under the unit interval linear DGP
model, OLS estimates of parameters are consistent and asymptotically normal under
general linear assumptions. However such assumptions are strong and is unlikely to
hold.
A common variant of the LPM for the fractional dependent variable (]0,1[) is the
logistic transformation y∗ = log(
y1−y
). However this variable is undefined for
observations at y = 1, and therefore not a viable approach.
86
TOBIT
The most common approach implemented in the current DEA literature is the two-
limit TOBIT regression framework. The two-limit model assumes that there is a
latent variable of interest y∗ where −∞ < y∗ <∞. The observable variable y is the
result of a censoring Data Generating Process (DGP) of y∗, where y = 0 if y∗ ≤ 0
and y = 1 if y∗ ≥ 1. More formally we describe the model using the following set of
equations:
y∗i =n∑k=1
βkxk + εi,
yi =
[1 + sign(y∗i )
2
]min {1, y∗i },
sign (y∗i ) =
1; y∗i ≥ 0
−1; y∗i < 0
where εi ∼ N(0, σ) are independent and identically normally distributed (iid.)
residual of the observations. The structural assumption of the error term allows
one to compute the densities of the TOBIT model.
The probability that a recorded y is equal to 0 is given by
P (y = 0) = F (y∗ ≤ 0)
= F (∑
βkxk + ε ≤ 0)
= F ( ≤ −∑
βkxk)
= F (ε− 0
σ≤ −
∑βkxk − 0
σ)
= Φ(−∑βkxkσ
)
(7.1)
Under this framework, the estimated parameters βkσ
are the effect of the exogenous
environmental variables on the index function inside the normal CDF (Φ). Also
note that βk is indirectly identified; the identification of this true parameter requires
87
positive density at the upper bound. In this case, the probability of y = 1
P (y = 1) = F (y∗ ≥ 1)
= 1− Φ(1−
∑βkxk
σ)
(7.2)
The estimation of 1σ
allows one to compute the underlying value of σ which can then
be utilised in identifying βk. Finally, the probability of y when it is between the
unit interval is given by:
P (yi|0 < yi < 1) = f(yi − βkxk)
= φ(yi −∑
βkxk)
(7.3)
where φ denotes the probability density function of the underlying normal distri-
bution of the error term. The joint likelihood function for the recorded censored
dataset is then given by:
L =∏yi=0
P (yi = 0) ·∏yi=1
P (yi = 1) ·∏
0<yi<1
P (yi|0 < yi < 1) (7.4)
The joint likelihood function is a major source of criticism of the two limit TOBIT
model, because the inherent nature of DEA efficiency prohibits a 0 efficiency score.
The absence of observations at 0 essentially eliminate the y = 0 component of
the joint likelihood function. The estimation is therefore based on a one limit
TOBIT model for y ∈ [−∞, 1]. Despite this, the said TOBIT model is still
valid in identification of the underlying parameters (i.e. β, σ) from the remaining
components of the likelihood. As such, Maddala (1986) has shown TOBIT to be an
sensible estimate of the conditional mean E(y|x).9
9See Appendix for full derivation
88
E(y|x) =∑
βkxk
[Φ
(1−
∑βkxk
σ
)− Φ
(−∑βkxkσ
)]+ σ
[φ
(−∑βkxkσ
)− φ
(1−
∑βkxk
σ
)]+
[1− Φ
(1−
∑βkxk
σ
)]The expression reveals a complex relationship between the conditional mean and
the index function. The corresponding marginal effects of the TOBIT model can
then be derived by taking the first order derivative of the above expression.
∂E(y|x)
∂xi= βi
[Φ
(1−
∑βkxk
σ
)− Φ
(−∑βkxkσ
)]
The marginal effect of covariate xi is a function of the coefficient estimateβi as well
as the other covariates and their respective coefficient estimates. More simply, the
marginal effect is the estimated coefficient βi scaled by the probability that the
conditional mean is within the bounds of [0,1]. This is an important result as it
indicates that if there are only handful observations that lie on the limiting values
of y, then the marginal effect will be very close to the estimated coefficient or the
LPM marginal effect.
A summary of the result using LPM and TOBIT specifications with robust standard
error and fixed year effect are presented below. Interactions between key explanatory
variables are also included to account for heterogeneous effects.10
10Only informative estimates are reported, for full regression output see appendix
89
Table 7.4: Efficiency Regression: LPM & TOBIT
(1) (2)LPM TOBIT
Corporate Fund Dummy −0.0179 −0.0211(0.0111) (0.0180)
Industry Fund Dummy −0.0470∗∗∗ −0.0565∗∗
(0.0172) (0.0240)Public Sector Fund Dummy −0.0607∗∗∗ −0.0641∗∗
(0.0202) (0.0257)Public Offer Dummy −0.0308∗∗∗ −0.0293∗∗∗
(0.00781) (0.0104)Defined Benefit Dummy −0.00396 0.00640
(0.0149) (0.0220)Corporate × Exit −0.0323∗∗ −0.0354∗∗
(0.0136) (0.0147)Corporate × Choice 0.0198∗∗∗ 0.0199∗∗∗
(0.00367) (0.00613)Public Sector × Choice −0.0341∗∗∗ −0.0379∗∗∗
(0.0101) (0.0141)Corporate × Share −0.120∗∗∗ −0.119∗∗∗
(0.0395) (0.0318)Public Sector × Share 0.0782∗∗∗ 0.0899∗∗∗
(0.0199) (0.0318)Exiting Market Dummy 0.0278∗∗ 0.0326∗∗
(0.0126) (0.0133)Market Share −0.0830∗∗∗ −0.0817∗∗∗
(0.0129) (0.0186)Assets in Default Option 0.0000870 0.000107
(0.0000782) (0.000113)Herfindahl Index −0.0321∗∗∗ −0.0315∗∗∗
(0.00694) (0.00694)Investment Choices (log) −0.0180∗∗∗ −0.0201∗∗∗
(0.00299) (0.00501)Constant 1.053∗∗∗ 1.062∗∗∗
(0.0162) (0.0209)
σ 0.124∗∗∗
(0.00589)
Adjusted R2 0.472Right Censored 7.3%Standard errors in parentheses∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Our results appears to be sensible and consistent with our hypotheses. Perhaps
the more striking result here is that after controlling for covariates, corporate funds
appears to be very similar to retail funds in their efficiency performance. Industry
and public sector super funds appear to be less efficient than retail and corporate
super funds. Further, our results show that market share has a significant and
negative impact on super fund efficiency. This is not a surprising result as our
90
previous results in this chapter suggested that structural features of the industry
could prevent highly efficient funds from expanding and allowing less efficient funds
to populate significant portions of the market. As expected, our result also shows
some heterogeneous effects of covariates among different fund types, where corporate
super funds are more susceptible to changes in market share when compared to
industry and retail funds. This is consistent with our a priori that the willingness
of employers to subsidise the cost borne by corporate funds is a decreasing function
of scale. Our estimates for the coefficients of other environmental variables are also
sensible.
Statistical inferences from both the TOBIT and LPM model rely heavily on
normality and the homoskedasticity assumption of the error term. McDonald (2009)
and Ramalho et al. (2010) argued against the implementation of TOBIT models
because the error term assumption is unlikely to hold due to its misspecification of
the underlying DGP for the variable of interest (i.e. efficiency scores are not results
of censoring or corner solutions). Similar criticism of LPM is also true. Pagan and
Vella (1989) conditional moment test and the Andrews (1988) Chi - square test
provides useful frameworks to assess the validity of these assumptions.
The misspecification under conventional linear models imply that the LPM/TOBIT
coefficients are likely to be inconsistent and thus biased. We then turn to more
sophisticated technique such as Fractional Response Models to produce more reliable
results.
Fractional Response Model
McDonald (2009) argued that the efficiency score data can be better described as
a normalisation process. In input oriented analyses, a production unit’s efficiency
score is determined as its actual inputs divided by the frontier inputs corresponding
to its output values. This normalises the maximum efficiency to 1 and all efficiency
91
scores to lie on or within the unit interval. To fully reflect this characteristic of
the underlying DGP we turn to the Fractional Response Model (FRM) proposed by
Papke and Wooldridge (1996).
The core assumption under the FRM is that the conditional mean of the response
variable is related to the linear index through a link function.
E(y|x) = G(xθ)
Where G(·) is a non linear function satisfying the constraints (0 ≤ G(·) ≤ 1). The
partial effect then follows
E(y|x)
∂xj= θjg(xθ)
Here we examine, four standard FRM GLM link functions under binomial dispersion:
G(xθ) =
exθ
1+exθLogit
Φ(xθ) Probit
ee−xθ
Loglog
1− ee−xθ Cloglog
Papke and Wooldridge (1996) have shown that regardless of the true distribution of y
conditional on x, as long as the link function is correctly specified, the estimator θ is
consistent and asymptotically normal. Figure 7.2 describes the cumulative density
G(xθ) and the pdf g(xθ) of the alternative models. An important observation is that
the maximum partial effect for the symmetric probit and logit models are achieved
at E(y|x) = 0.5 and are identical for values of x that yield values of E(y|x) that
are symmetric around that point (Ramalho et al., 2010) That is, the effect of x on
E(y|x) is the same for E(y|x) = 0.05 and E(y|x) = 0.95.
While asymmetric loglog and cloglog yields a varying partial effect of x at different
values of E(y|x) it can be observed that the models are equivalent to each other at
E(y|x) = 0.5. Further, under asymmetric models, x achieves the highest partial
92
Figure 7.3: Standard Fractional Regression Models
Source: Ramalho et al. (2010)
effect at values E(y|x) < 0.5, E(y|x) > 0.5. These features of the alternative
specifications should reflect the observed characteristics in the data, and thus a
misspecification test is conducted to assess the underlying specification of the model.
The Quasi Maximum Likelihood Estimation (QMLE) procedure in Papke and
Wooldridge (1996) follows the Bernoulli log-likelihood function
Li(θ) = yilog [G(xiθ)] + (1− yi)log [1−G(xiθ)]
A summary of the results are presented in Table 7.6, where we also apply fixed year
effect and clustered standard errors.11
11For full regression output see appendix
93
Table 7.5: FRM Results with Four Link Function Specifications
(1) (2) (3) (4)
Logit Probit Loglog Cloglog
Corporate Fund Dummy 0.438∗∗∗ 0.165∗∗ 0.459∗∗∗ 0.0824∗
(0.145) (0.0696) (0.137) (0.0487)
Industry Fund Dummy −0.596∗∗∗ −0.288∗∗∗ −0.569∗∗∗ −0.203∗∗∗
(0.177) (0.0911) (0.165) (0.0668)
Public Sector Fund Dummy −0.674∗∗∗ −0.330∗∗∗ −0.629∗∗∗ −0.239∗∗∗
(0.240) (0.123) (0.214) (0.0926)
Public Offer Dummy −0.314∗∗∗ −0.151∗∗∗ −0.286∗∗∗ −0.104∗∗∗
(0.0764) (0.0388) (0.0714) (0.0286)
Public Sector × Choice −0.243∗∗ −0.144∗∗ −0.194∗∗ −0.131∗∗∗
(0.108) (0.0567) (0.0916) (0.0444)
Corporate × Share −1.477∗∗∗ −0.796∗∗∗ −1.335∗∗∗ −0.634∗∗∗
(0.218) (0.131) (0.139) (0.147)
Public Sector × Share 0.423∗∗∗ 0.258∗∗∗ 0.297∗∗∗ 0.291∗∗∗
(0.119) (0.0699) (0.0890) (0.0726)
Exiting Market Dummy 0.643∗∗ 0.272∗∗ 0.647∗∗∗ 0.163∗∗
(0.257) (0.119) (0.243) (0.0795)
Market Share −0.467∗∗∗ −0.276∗∗∗ −0.351∗∗∗ −0.297∗∗∗
(0.0676) (0.0383) (0.0333) (0.0488)
Herfindahl Index −0.365∗∗∗ −0.185∗∗∗ −0.345∗∗∗ −0.133∗∗∗
(0.0670) (0.0342) (0.0622) (0.0255)
Investment Choices (log) −0.182∗∗∗ −0.0952∗∗∗ −0.173∗∗∗ −0.0690∗∗∗
(0.0272) (0.0139) (0.0238) (0.0104)
Constant 3.891∗∗∗ 2.139∗∗∗ 3.825∗∗∗ 1.472∗∗∗
(0.184) (0.0897) (0.173) (0.0634)
Log-likelihood −524.8 −527.4 −523.3 −530.9
AIC 1099.7 1104.9 1096.5 1111.9
BIC 1245.4 1250.6 1242.3 1257.6
Standard errors in parentheses
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Our estimated coefficients represent the expected contribution of the covariates on
the index function instead of the dependent variable of interest. The effects are
also dependent on the levels of other covariates, therefore before any meaningful
interpretation can be made, we must assess the model specification for each of the
link functions considered and compute the corresponding marginal effects under that
specification.
94
Misspecification
As previously stated, the FRM only yields a consistent estimator θ when the
functional form of the link is correctly specified. One way to assess the general
specification errors of the alternative models is the Ramsey RESET test. Pagan
and Vella (1989) showed that any index model of the form E(y|x) = G(xθ) can
be approximated by S(xθ +∑J
j=1 γj(xθ)j+1) for sufficiently large J . We follow this
logic and adopt a two stage process, the first stage procedure regresses the efficiency
scores against all covariates using FRMs; the second stage adopt identical procedure
but also incorporate the quadratic and cubic terms of the predicted values in first
stage. We then assess the significant of the quadratic and cubic term. The RESET
test results are presented in Table 7.7.
Table 7.6: RESET Test Results
(1) (2) (3) (4) (5) (6)
OLS TOBIT Logit Probit Loglog Cloglog
y2 t-test 29.62∗ 15.78 0.381 −0.379 −0.291 −1.203
(16.52) (18.40) (6.372) (4.079) (4.664) (4.006)
y3 t-test −12.95∗∗ −7.614 1.613 2.142 1.730 3.195
(6.171) (6.786) (5.757) (3.593) (4.372) (3.450)
y2 & y3 f-test 17.83∗∗∗ 17.10∗∗∗ 4.30 10.14∗∗ 3.08 15.82∗∗∗
(p− value) (0.00) (0.00) (0.12) (0.01) (0.21) (0.00)
Standard errors in parentheses
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
All models except for Logit and Loglog are rejected at a 10% significance level for
the joint significance test. That is, the predicted terms are not significantly different
from 0, and thus provide no explanatory power to the conditional mean. To further
inform model selection, we examine the relative fit of the two models.
Model Fit
A further analyses of model fit is conducted using Akaike Information Criterion
(AIC) and Baysian Information Criterion (BIC). These measures provide a means
to test the statistical quality of the models through their likelihood.
95
Table 7.7: Information Criterion
Logit Probit Loglog Cloglog
AIC 1099.7 1104.9 1096.5 1111.9
BIC 1245.4 1250.6 1242.3 1257.6
AIC is theoretically motivated and assesses statistical fit of the model using
the likelihood estimates, BIC employs a similar framework but punishes less
parsimonious models. However, in this case, all four link functions have the same
number of estimated parameters, therefore the AIC is sufficient for making model
selection. According to AIC measures, loglog has the best relative fit, while logit
ranked second.12
These test result suggest that loglog link is the sensible model selection. Future
results discussed in the rest of this chapter will reflect this model selection.
Partial Effects
The partial effect of the FRM with loglog Link can be seen from the previous
generalisation
E(y|x)
∂xj= θjg(xθ) = θje
xθ−exθ
It is dependent on the coefficient estimate of xj as well as the value of the index
function, that is, the magnitude of the partial effect is dependent on the value
of other covariates. To generalise inference of the partial effects, two measures are
computed: (1) Partial effect At the Means (PAM); (2) Average Partial Effect (APE).
Results are presented in Table 7.9.
12The less the better
96
Table 7.8: FRM (loglog) Partial Effects
(1) (2)
Partial Effect At Mean Average Partial Effect
Corporate Fund Dummy 0.0286∗∗∗ 0.0407∗∗∗
(0.00866) (0.0123)
Industry Fund Dummy −0.0355∗∗∗ −0.0505∗∗∗
(0.0102) (0.0146)
Public Sector Fund Dummy −0.0392∗∗∗ −0.0557∗∗∗
(0.0132) (0.0189)
Public Offer Dummy −0.0178∗∗∗ −0.0254∗∗∗
(0.00442) (0.00631)
Corporate × Share −0.0832∗∗∗ −0.118∗∗∗
(0.00893) (0.0126)
Public Sector × Share 0.0185∗∗∗ 0.0263∗∗∗
(0.00558) (0.00786)
Industry × Share 0.00652∗ 0.00927∗
(0.00388) (0.00549)
Public Sector × Choice −0.0121∗∗ −0.0172∗∗
(0.00572) (0.00810)
Corporate × Exit −0.0185 −0.0263
(0.0175) (0.0250)
Exiting Market Dummy 0.0403∗∗∗ 0.0573∗∗∗
(0.0149) (0.0215)
Market Share −0.0219∗∗∗ −0.0311∗∗∗
(0.00220) (0.00296)
Assets in Default Option 0.000108∗ 0.000153∗
(0.0000561) (0.0000798)
Herfindahl Index −0.0215∗∗∗ −0.0306∗∗∗
(0.00394) (0.00560)
Investment Choices (log) −0.0108∗∗∗ −0.0154∗∗∗
(0.00145) (0.00209)
Observations 2512 2512
Standard errors in parentheses
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Our result shows that PAM estimates are consistently smaller in scale than that of
APE. This is consistent with the concave curvature of partial effect g(xθ), where the
APE is the average of g(xθ) and PAM is a line representing the linear combination of
two points on the curve, concavity ensures that APE ≥ PAM . While both measures
97
are valid and their interpretations are slightly different. PAM is an estimate of the
partial effect at the mean value of the covariates, generalised inference using PAM
is valid if majority of covariates are continuous. In our research, most exogenous
contextual variables are binary dummies, the mean values of these variable are not
achievable in reality, this then limits our ability to inference based on these results.
APE represent the mean of the partial effect over the entire sample. It has a simple
intuitive interpretation as the expected value of the partial effect. Inference based
on APE is consistent if the sample covariate distribution is representative of the
population. This is true in our case since the data set captures all regulated funds
under APRA’s regulatory jurisdiction, with only small sample truncations based on
erroneous data. Due to these considerations, interpretation of our regression result
is based on the APE of the FRM using loglog link.
FRM partial effects closely resembles those results under conventional linear models.
This is not surprising as conventional linear models are often good approximations
of partial effects averages or at the mean value of the index function. The true
advantage of these FRM consistent estimators is in computing the partial effects
at the end points, where linear model prediction introduce significant bias due to
poor specification of the underlying data structure. The result is analogous to
that of McDonald (2009), where the paper recognised model misspecification under
LPM/TOBIT but argued that for the purposes of most studies, OLS is sufficient
in approximating the true APE. Regardless, this section of this research serves to
ensure that our result is robust under the various model specifications, and that any
inference based on these results are valid.
7.3.3 Discussion of Results
Our second stage DEA regression results presents a number of implication for the
efficiency and productivity of Australian superannuation funds. In particular, our
results presents quantifiable evidence of the influence of the various structural and
design features in the Australian superannuation environment.
98
Fund Type
Our second stage result is consistent with our previous result which showed that
corporate super fund appear to be the most technically efficient fund type in the
market. Relative to retail fund, corporate funds on average outperform by a PTE
score of 0.0407. Again, this result is not surprising, as our analysis incorporates cost
subsidization of the corporate funds as input discounts. This allows corporate funds
to operate at a relatively lower cost base, which could explain how super funds in
this sector is able to systematically outperform the industry in technical productive
efficiency.
Perhaps the more interesting result here is that, after controlling for other
environmental factors, we found that retail funds are more efficient than public
sector and industry funds. This result is at odds with recent Australian literature
on superannuation fund investment performance and administrative fees, which have
generally concluded that retail funds appear to charge members higher fees but offer
very little difference in investment outcomes (Bateman and Mitchell, 2001; Coleman
et al., 2003; Ellis et al., 2008). The difference in our results can be explained in two
ways.
First, previous studies have focused on financial metrics such as investment outcomes
and administrative expenses. In additional to these traditional measures of
performance, our analysis has also accounted for the membership characteristics
of the funds by adopting the VaR output measure. This measure punishes super
funds for excessive risk exposure for shorter term asset holdings (i.e. those members
nearing retirement), which could help explain why retail super funds appear to
be superior in their technical efficiency to deliver this output. Retail super funds
by their nature are a client winning business and typically experience greater
99
membership engagement. We can then hypothesise that portfolio holdings in retail
funds are much more tailored to the needs of members when compared to other
fund types. This could be a result of active management by the fund managers, or
deliberate life cycle choice by the members. Unfortunately the limitations of our
data does not allow the separate identification of these two dynamic movements in
portfolio composition. To the best of our knowledge, this research is among the
first to incorporate membership characteristics in the assessment of superannuation
fund performance in Australia. In particular, the incorporation of member age
profiles and the corresponding fixed statutory asset holding period they face in our
performance metric provides a better description of the industry’s core functionality
in the provision of retirement incomes.
Second, literature that studies the performance differentials between different fund
types have rarely controlled for other structural characteristics of the super funds,
including features such as market share, investment choice, market concentration,
public offer etc. Our results are more robust in that technical efficiency comparisons
between super fund types are conducted under a regression framework, where we
have controlled for environmental variables that could potentially contribute to the
technical efficiency of the super funds. This could also contribute to the difference
between previous results and our results.
Public Offer
Recall that the public offer dummy captures whether the super fund offers its services
to the general public. It is also a proxy variable that characterises the trustee board
structure. Our results shows that public offer status has a negative effect on the
technical efficiency of super funds; on average a public offer super fund is likely to
be 0.0178 less efficient than its counterpart.
Again, this result is consistent with our a priori that public offer funds are likely
100
to incur higher administrative expenses due to expenditure in areas including
marketing, distribution and other related areas.
The findings of the Australian literature in this respect is mixed. Drew and Stanford
(2003) argue that public offering allows for more vigorous price competition and
promotes market discipline of funds, so under such assumption we would likely to
see public offer funds outperforming industry counter parts in efficiency. However
Vidler (2004) argued that such an approach ignores cost issues inherent in personal
account based pension systems, and costs associated with competition observed both
in Australia and overseas. The evidence presented in our research seems to support
the latter, where competition in the general superannuation market appears to have
a negative impact on the super funds’ technical productive efficiency.
Market Power
Our results showed that on average, a 1% increase in market share will result in
a 0.0311 decrease in technical efficiency for industry and retail funds. Inclusion
of interaction terms also revealed significant heterogeneous effects, where corporate
fund suffer an additional 0.118 efficiency loss from gains in market share; while
public sector funds suffer slightly less than industry and retail funds by 0.0263. The
net effects are presented in Table 7.10
Table 7.9: Net APE of 1% change in market share
Corporate Retail Industry Public Sector
1% ∆ Market Share -0.1491 -0.0311 -0.0311 -0.0048
Overall, our results show that increases in market share systematically decrease
productive efficiency across all sectors of the industry. However, this is does not
necessarily mean that there are no efficiency gains from economies of scale. Rather,
we observe the overall effect of the trade-off between costs arising from market
power and production economies of scale benefits. The negative effect of market
101
power would simply suggest that the cost is greater than the benefit gained.
The results illustrate that corporate super funds suffer the most from changes in
market share. This is consistent with our a priori; recall that corporate funds enjoys
significant input factor discounts as a result of cost subsidisation from the sponsoring
employer. This is also a caveat that limits the growth of corporate super funds, as
the real cost of borne by the employer to subsidise the operations of the corporate
fund is directly related to the relative scale of the fund. Increasing market share,
or fund scale is likely to decrease the employers’ ability and willingness to continue
supporting the super fund’s operations. In essence the loss of employer subsidies
removes any competitive cost advantage of corporate super funds, and is likely to
result in large decreases in technical efficiency.
Both public offer super funds, industry13 and retail sectors appears to suffer similar
efficiency losses from increases in market power. This could be attributed to the costs
arising from significant gains in market share. Vidler (2004) suggested that in order
to gain significant market power in a competitive environment, super funds have to
offer a greater range of products to accommodate greater number of members, invest
in new IT systems to cope with increasing level of member interaction, and most
importantly, develop and maintain marketing exposure to compete against industry
counterparts.
As expected, public sector funds suffer from neither caveats (cost subsidies, or
competition) and therefore remain relatively immune to market share movements.
Investment Options
Our results suggest that more investment choices offered by the fund decreases fund
efficiency: a 1% increase in investment choices on average decreases super fund
13Not all industry funds are public offer, but increasing number of industry funds are undergoingsuch reform.
102
efficiency by 0.0154. This effect is also heterogeneous for public sector funds, where
the sector experiences an additional 0.0172 efficiency loss from increased investment
choices.
Our result is intuitive and consistent with the majority of recent industry studies.
Bateman and Thorp (2007) argued that inefficiency in decentralized portfolio
management can arise from incomplete information transfer between the central
manager and delegated manager. This is difficult, and inefficiency is likely to
arise from unsuccessfully combining delegated portfolio subsets into an efficiency
centralized portfolio. Further, increasing investment options offered to members
is also likely to increase fixed portfolio management costs, hence decrease overall
productive efficiency.
It is unclear why public sector super funds experience heterogeneous effects from
this characteristic. One plausible explanation is perhaps that these funds have
significantly higher cost base in administering additional investment options. There
is a lack of evidence that is able to provide any explanation to this effect. Perhaps
this could be resolved in future research.
Market Concentration
Recall that the Herfindahl index measures the level of competition and market
concentration in each year of our sample. Our result illustrate that this variable
is negatively related to the fund efficiency scores. A 0.01 movements in the index
correspond to a universal efficiency loss of 0.0306 across the industry. This is a
sensible result given our previous analysis of market shares, where increasing fund
scale decreases its productive efficiency. Although the Herfindahl index is somewhat
collinear to the market share variable, they are separately identifiable.
One should also note that the current Herfindahl index of the Australian superan-
103
nuation market is relatively low, with market share largely decentralised. Therefore
it is unlikely we would see significant changes in the index over the next few years.
This implies that the variable is unlikely to have a significant impact in the short
term.
Wind Up Funds (Exits)
Our previous analysis in the first stage revealed that market leavers in the industry
appears to be more efficient than incumbent funds. This holds true in our second
stage results, where market leavers are 0.0573 more efficient on average. Corporate
funds exits also appears to be heterogeneous, where the effect is reduced by 0.0263
to 0.0310.
Similar to our first stage analysis, these results clearly violate the efficient market
hypothesis. We have attributed this effect to two structural failures in the market.
First, the market structure is restrictive and segregated where some super funds
only compete on a sub population level. This limits the ability of efficient fund to
expand their outputs and increase their market power. Second, some super funds
derive their efficiency from competitive cost advantage due to subsidies from fund
sponsors. These subsidies are not sustainable in the long run as funds increase in
scale. These factors remains valid in our second stage interpretation of results.
Recall, one of the key motivations of this research was to better understand the
persistence of the average expense to asset ratio of the industry over the past decade.
This could be partly be explained by our result relating to market exits. Where
super funds that exit the market appear to be more efficient than their incumbent
counterparts, this could be interpreted as a downward shift in the frontier or the
efficiency cluster below the frontier. In practice, less efficient funds require more
inputs to produce the same level of output, therefore asset flows from more efficient
funds to less efficient funds within the industry are likely to counteract any cost
104
reductions arising from increases in economies of scale due to consolidation and
growth.
7.4 Policy Implication
These results have significant implications for future policy formulation and
regulatory reforms.
We find that corporate funds are systematically more efficient than the rest of the
industry. Although these efficiencies can be partially attributed to discounted inputs
factors due employer subsidies, they benefit the members nonetheless. Therefore,
future superannuation policies should encourage the continuum of corporate fund in
the market. This can be achieved by reducing excess fixed regulatory costs associated
with licensing and compliance for smaller funds.
Our result also shows that the costs associated with public offer is likely to outweigh
any benefit derived from the reduction in prices. This is consistent with the finding
of the Cooper Review, which suggested cost transparency as a major impediment
to reducing costs. The limited visibility is likely to hinder any vigorous pricing
competition among competitors, hence offering little incentive for super fund to
reduce the underlying cost to deliver services.
This research also find that increasing market power is likely to decrease efficiency.
This seems to provide some empirical evidence to Vidler (2004), which argued that
in order to gain significant market power in a competitive environment, super
funds must allocate significant resources in improving their market profile and
accommodate more members. This results calls for more restrictive anti-competitive
regulation, where super fund mergers and acquisitions should be discouraged.
We also find that large number of investment options decrease super fund efficiency.
105
This provides empirical support to the recent policy reforms to mandate default
options and consolidate the dispersion of portfolio choices.14
7.4.1 Extension: Aggregate Performance
Efficiency measurements under the Data Envelopment Analysis framework are
computed using a piece wise comparison mechanism. A common criticism of this
technique is its inability to account for standard errors. For example, an inefficient
super fund could be considered efficient for period t if their particular portfolio
worked well with the unexpected financial shocks during that period. Indeed, our
first stage analysis revealed that selected funds significantly outperformed the rest
of the industry during the period of the GFC. The superior performance of these
super funds could reflect their true productive efficiency or could be attributed a
degree of luck. Relatively better performance of super funds such as BMA Personal
in 2008/2009 could be traced to their significant holding of defensive assets such as
domestic bonds and cash. These asset categories are less risky during periods of
market downturn, hence assist in limiting losses incurred by the super funds during
these periods. However, these portfolios are not necessarily the best investment
strategies in other periods as they offer relatively lower levels of return. The
objective of this extension is to ensure that our results are robust under the aggregate
performance framework. This section will briefly discuss the data generating process
underpinning this analysis and compare the results with our previous findings. The
interpretation is brief and is analogous to that of the previous section.
Data
To compute aggregate performance, we take the average of super funds input/output
measures over the sample period from 2004-2012. DEA analysis is then computed
based on these observations.
The unbalanced nature of our data presents a unique problem. Super funds that
14MySuper reform 2012
106
exit the market before the financial crisis are likely to have higher output measures,
which would give them an unfair advantage in the DEA process. Therefore we also
check the robustness of our results using a sample comprising super funds which was
present in the data for at least 5 years.15 However, analysis under the restrictive
sample could also cause sample selection issues, as attrition is asymmetric and biased
towards corporate and retail funds.
Results
Table 7.10: Average Efficiency Scores: Aggregate Performance
Full Sample Restricted Sample
Average PTE Std. Dev Obs Average PTE Std. Dev Obs
Corporate 0.957 0.079 296 0.966 0.039 103
Industry 0.748 0.196 78 0.870 0.095 67
Public Sector 0.747 0.216 25 0.859 0.097 18
Retail 0.822 0.221 183 0.843 0.179 108
Overall 0.878 0.181 574 0.892 0.134 288
The first stage results presented in Table 7.11 are similar to our previous findings.
Corporate super funds lead the industry in average productive efficiency, followed
by retail super funds. An interesting result here is that under the restricted sample,
retail funds did not appear to be significantly different from industry and public
sector super funds. In fact, we observe a systematic increase in average efficiency
across all four sectors. This could be partly explained by our sample selection
strategy. By restricting the sample to super funds that lasted for more than 5 years,
we reduce the amount of distortion created by market leavers. This has resulted
in a lower dispersion of efficiency score (lower Std.Dev); particularly in the lower
spectrum, hence boosting efficiency averages.
15Due to lack of new market entrants, only super fund that exits before 2009 are dropped
107
We employ a similar FRM estimation technique in our previous second stage
analysis. Specification tests and model fit suggests that Cloglog would be the correct
link function in this instance. A summary of our results are presented in Table 7.12
Table 7.11: FRM (cloglog) Partial Effects: Aggregate Performance
(1) (2)Full Sample Restricted Sample
Corporate Fund Dummy 0.00166 0.0184(0.0197) (0.0257)
Industry Fund Dummy −0.0483∗ −0.0184(0.0289) (0.0351)
Public Sector Fund Dummy −0.0613 0.00761(0.0394) (0.0372)
Public Offer Dummy −0.0269∗∗ −0.0306∗∗∗
(0.0108) (0.0109)Corporate × Share −0.134∗ −0.0259
(0.0745) (0.0401)Public Sector × Share 0.175∗∗∗ 0.0507∗
(0.0413) (0.0283)Industry × Share 0.110∗∗∗ 0.0707∗∗
(0.0415) (0.0293)Public Sector × Choice −0.0365 −0.0343∗
(0.0245) (0.0203)Corporate × Exit 0.00253 0.0109
(0.0164) (0.0233)Exiting Market Dummy 0.0598∗∗∗ 0.0121
(0.0130) (0.0191)Market Share −0.161∗∗∗ −0.0714∗∗∗
(0.0334) (0.0248)Assets in Default Option 0.000192 0.000149
(0.000135) (0.000225)Investment Choices (log) −0.0146∗∗∗ −0.0148∗∗
(0.00424) (0.00616)
Standard errors in parentheses∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
Under the full sample analysis, we find that there is little difference in efficiency
performance between fund types except for industry funds, which slightly under
performs the other three fund types after controlling for other covariates. Although
this result differs from our previous findings, this is not entirely unexpected. Our
previous analysis has shown that retail funds are only relatively more productive
108
than other fund types during market downturns, so it is plausible that this
identification is no longer viable under the aggregate performance framework.
Consistent with our previous findings, public offering appear to have a negative
relationship with super fund efficiency. Our estimated effects of public offering are
fairly robust, with the effect under the two aggregate samples (-0.0269 and -0.0306
respectively) only slightly higher than the original APE estimates (-0.0254). This
result confirms that public offer super funds are generally less efficient than their
non-public offer counterparts. Similar findings also extend to investment choices,
with the estimated effect fairly similar under the three sample specifications (-0.0154,
-0.146 and -0.148).
The results in Table 7.12 confirms our previous finding in relation to the effect of
market power on super fund efficiency. We find a consistent negative relationship
across the two aggregate samples (-0.161, -0.0714). The difference in the estimated
effect between the two samples could be attributed to the attrition of small market
exit super fund16 in the restricted sample. Regardless, these estimates appear to
be significantly higher than our original estimate of -0.0311. However, one should
keep in mind that the market share variable regressed here is the average measure
over the entire period that the super fund appeared in the sample. Therefore, this
estimate differs from our original results as the dynamic movements in market power
are no longer captured under the aggregate samples. In addition, we find similar
heterogeneous effect across fund types, with corporate super funds experiencing the
higher marginal effect, followed by industry and retail funds.
Perhaps the more interesting result here is our estimated effect for market exits.
Under the full sample specification, the effect of a market exit is 0.0598. This
is nearly identical to our previous APE estimate of 0.0573. In contrast, we find
16Market exits before 2009 are comprised of mainly small corporate funds
109
no significant differences between market exits and incumbent players under the
restricted sample analysis. This a surprising result, as the restricted sample only
captures market exits from 2010-2012, and suggest that market exits are randomized
in terms of productive efficiency. However, it is important to note that there is a
very limited number of market exits between 2010-2012, therefore there may be a
small sample bias for the estimator due insufficient observations.
Overall, our results are robust under the aggregate performance measures. This
implies that inferences based on our results are valid despite the possible ’luck’
element under the DEA framework.
7.4.2 Key Results
This section will provide a summary of the key results presented in this chapter. In
general, we find that:
• The Australian superannuation industry appears to have relatively high
technical and scale efficiency. However, the industry could benefit from
improvements in scale efficiency of public offer super funds.
• Corporate super funds consistently appear to be more efficient than their
industry counterparts. As well, the majority of the highly efficient retail funds
are affiliated to commercial banks.
• On average, industry, public sector and retail super funds appear to be equally
efficient. However, retail funds appear to perform slightly better during market
downturns.
• Conventional linear regression frameworks are inadequate for specifying the
underlying data generating process of the efficiency scores. Fractional response
models using binomial dispersion and a loglog link function appear to provide
consistent estimates of environmental variables.
110
• After controlling for contextual variables, we find that corporate funds are the
most efficient fund type, followed by retail and industry super funds. Public
sector super funds appears to be the least efficient.
• Super funds that offer their membership to the general public are less efficient
than their not-for-profit counterparts.
• Market Power is negatively related to the technical efficiency of super funds.
A heterogeneous treatment effect is also present for different fund types, with
corporate funds suffering highest declines with market share gains.
• The number of investment options decreases technical efficiency of super funds.
• Centralisation of portfolio assets in the default option increase technical
efficiency of super fund, although the effect is not economically significant.
• Defined Benefit design did not have an impact on super fund efficiency.
The above results are also robust under an aggregated performance framework,
where efficiency is computed using average measures across the sampling period.
111
Chapter 8
Conclusion
The aim of this research was to investigate the production efficiency of the Australian
superannuation industry, and discover the relative impact of key industry structural
and design features. The efficiency analysis was conducted using a two stage
construct. In the first stage a non-parametric frontier technique - Data Envelopment
Analysis (DEA) - was calibrated to measure the production efficiency for a sample of
APRA regulated large Australian superannuation funds between 2004 and 2013. In
the second stage, technical efficiency of super funds are analysed under a regression
framework, where the impact of key industry structural characteristics and design
features were assessed.
A review of the international literature revealed significant insights from efficiency
studies on the banking sector and mutual funds. These research have helped to
inform government policy formulation and regulatory reforms. By contrast, studies
of superannuation fund efficiency in the Australian literature is very limited. The
core contribution of this research is to bridge this literature gap, and then inform
policy and regulatory development and industry practice.
In comparison to other parametric efficiency analysis, DEA is superior within the
confines of our research as it does not require the input factor prices that are
not available for the current study. Further, the technique allows for multiple
output/input definitions of the production process. This is advantageous as it
enables us to capture the multiple functionality of the superannuation industry
and produce an aggregate measure of production efficiency based on their respective
outcomes. A key contribution of this research is to assess super fund performance
using a innovative Value at Risk framework, which readily incorporates membership
112
characteristics in evaluating super funds’ performance in risk management.
The majority of previous two stage efficiency studies have employed conventional
linear models in their analysis. These models are heavily criticised for their
misspecification of the underlying data structure and generally yield inconsistent
results. This research has assessed the validity of these models against the more
sophisticated non-linear modelling techniques. We found that the fractional response
model with binomial dispersion and loglog link function is the most accurate model
for correctly specifying the underlying data.
Our results show that the Australian superannuation industry has relatively high
technical and scale efficiency, with most super funds operating at or close to the
production frontier. We found that corporate super funds consistent outperform
their industry counterparts in technical efficiency, which is likely driven by the
discounted input factors resulting from heavy cost subsidization by corporate
sponsors.
This research also found that super funds that offer their membership to the general
public are less efficient than closed funds (i.e. non-public offer). This result has
significant policy implications, as it seems to discourage recent regulatory movements
to promote greater consumer choice in the selection super funds. Our results provide
some empirical basis for previous studies such as Vidler (2004), which argues there
are significant costs associated with increasing levels of competition and the cost
would outweigh any benefit derived from market discipline.
Another key finding of this research is the negative relationship between a super
funds efficiency and their corresponding market share. This result suggests that
any efficiency gains from increasing economies of scale is likely to be offset by
the increasing costs arising from gains in market power. The effect also varies
113
in severity across the different fund types, with corporate fund experiencing the
greatest efficiency loss, followed by retail and industry funds. The implications are
complex, and our findings suggest that the superannuation fund market is highly
inefficient, with less efficient entities occupying larger portions of the market. That
being said, we found that the more efficient super funds often derive their efficiency
from subsidies and are restricted to sub populations of the market, this essentially
limits their ability to expand. Such market restriction prevents any allocative gains
from reallocation of output between more efficient and less efficient super funds.
Finally, our results have shown that the majority market exits in the past decade
are highly efficient super funds. Previous studies have attributed corporate super
fund exits to the increasing cost of regulation and compliance. Our finding imply
that the current regulatory framework is inadequate in promoting efficiencies within
the market. In particular, fixed regulatory expenditures such as licensing costs and
the recent MySuper reforms serves to discourage small and highly efficient corporate
funds from participating in the market.
114
Bibliography
Andrews, D, W. K. (1988). Chi-square diagnostic tests for econometric models:
Theory. Econometrica 56, 1419 – 1453.
APRA (1998). Regulation: The new apra. Insurance and Superannuation Bulletin.
APRA (2005). Classification of superannuation entities. APRA Statistics .
Balk, B. M. (2001). Scale efficiency and productivity change. Journal of Productivity
Analysis 15, 159 – 183.
Banker, R. D., A. Charnes, and W. W. Cooper (1984). Some models for estimating
technical and scale inefficiencies. Management Science 30, 1078 – 1092.
Barrientos, A. and A. Boussofiane (2005). How efficient are pension fund managers
in chile. Revista de Economia Contempornea 9, 311.
Barros, C. P., G. Ferro, and C. Romero (2008). Technical efficiency and heterogeneity
of argentina pension funds. Technical University of Lisbon: Working Paper .
Barros, C. P. and M. T. M. Garcia (2006). Performance evaluation of pension funds
management companies with data envelopment analysis. Risk Management and
Insurance Review 9, 165.
Basso, A. and S. Funari (2001). A data envelopment analysis approach to measure
the mutual fund performance. European Jounral of Operational Research 135, 477
– 492.
115
Basso, A. and S. Funari (2003). Measuring the performance of ethical mutual funds:
a dea approach. European Journal of Operational Research 54, 521 – 531.
Bateman, H. (2001). Disclosure of superannuation fees and charges. Centre for
Pensions and Superannuation Discussion Paper 3(04).
Bateman, H. (2002). Retirement income strategy in australia. Economic Analysis
and Policy 32, 49.
Bateman, H., K. Geofferey, and J. Piggot (2001). Forced Saving: Mandating Private
Retirement Incomes. Oxford University Press.
Bateman, H. and O. Mitchell (2001). A new look at pension plan adminstrative
expenses. NBER Program on the Economics of Aging .
Bateman, H. and S. Thorp (2007). Decentralized investment management: an
analysis of non-profit pension funds. Journal of Pension Economics and Finance 6
(01), 21 – 44.
Berger, A., W. Hunter, and S. Timme (1993). The efficiency of financial institutions:
a review and preview of reserach past, present and future. Jounral of Banking
and Finance 17, 221 – 249.
Berger, A. and J. Mester, L (1997). Inside the blackbox: What explains differences
in efficiencies of financial instituions? Journal of Banking and Finance 21, 895 –
947.
Berger, A, N. and B. Humphrey, D (1997). Efficiency of financial institutions:
International survey and directions for future research. European Jounral of
Operational Research 98(2), 175 – 212.
Bhagavath, V. (2006). Technical efficiency measurement by data envelopment
analysis: an application in transportation. Alliance Journal of Business
Research 2 (1), 60–72.
116
Bird, R., H. Chin, and M. McCrae (1983). The performance of australia
superannuation funds. Australian Journal of Management 8, 49 – 69.
Braberman, D, G., L. Chisari, L. Quesada, and A. Rossi, M (1999). Funcion de costos
en la industria de las administradoras de fondos de jubilaciones y pensiones en
la argentina: Un analisis de las modificaciones regulatorias. Revista de Analisis
Economico 14, 121 – 142.
Charnes, A., W. W. Cooper, and E. Rhodes (1978). Measuring the efficiency of
decision making units. European Jounral of Operational Research 2, 429 444.
Coelli, T. J., C. J. O’Donnell, D. S. Prasada Rao, and G. E. Battese (2005). An
Introduction to Efficiency and Productivity Analysis. Springer.
Coleman, A., N. Esho, and M. Wong (2003). The investment performance of
australian superannuation funds. Australian Prudential Regulation Authority:
Working Paper .
Coleman, A., N. Esho, and M. Wong (2006). The impact of agency costs on
the investment performances of australian pension funds. Journal of Pension
Economics and Finance 5, 299 – 324.
Dowd, K., D. Blake, and A. Cairns (2004). Long term value at risk. Journal of Risk
Finance 5(2), 52–57.
Drew, M. and J. Stanford (2001). Asset selection and superannuation fund
performance: A note for trustees. Economic Papers: A journal of applied
economics and policy 20, 57–65.
Drew, M. and J. Stanford (2003). Returns from investing in australian equity
superannuation funds, 1991 - 1999. The Service Industries Journal 23, 12.
Ellis, K., A. Tobin, and B. Tracey (2008). Investment performance, asset allocation,
and expenses of large superannuation funds. Australian Prudential Regulation
Authority: Working Paper .
117
Fare, R., S. Grosskopf, and P. Roos (1998). Malmquist Productivity Indexes: A
survey of Theory and Practice. Boston: Kluwer Academic Publishers.
Farrell, M. (1957). The measurement of productive efficiency. Journal of the Royal
Statistical Society 120, 253 – 290.
Ferrier, G. D. and C. A. Lovell (1990). Measuring cost efficiency in bankingl
econometric and linear programming evidence. Journal of Econometrics 46, 229
– 245.
Galagedera, D. and P. Silvapulle (2002). Australian mutual fund performance
appraisal using data envelopment analysis. Managerial Finance 28 (9), 60–73.
Goddard, E., A. Weersub, K. Chen, and G. Turvey, C (1993). Economics of strutural
change in agriculture. Canadian Jounral of Agricultural Economics 41(4), 475–
489.
Grosskopt, S. (1993). The Measurement of Productive Efficiency - Techniques and
Applications. New York: Oxford University Press.
Humphrey, D. B. (1985). Cost and Scale Economies in Bank Intermediation, pp.
745 – 783. New York: John Wiley and Sons.
Jacob, N. (1998). Evaluating investment performance. New York: John Wiley and
Sons, Inc.
Kewley, T. H. (1973). Social security in australia, 1900-72. Sydney University Press,
Sydney , 3.
Kolari, J. and A. Zardkoohi (1987). Bank Cost, Structure and Performance.
Lexington, KY: Lexington Books.
Kumbhakar, S, C., E. Tsinas, and T. Sipilainen (2009). Joint estimation
of technology choice and technical efficiency: an application to organic and
conventional dairy farming. Journal of Productivity Analysis 31(3), 151–161.
118
Liu, K. (2013). Australian Superannuation: Operation Structure, Investment
Performance and Trustee Governance. Ph. D. thesis, University of Sydney.
Maddala, G, S. (1986). Limited-dependent and Qualitative Variables in
Econometrics. Cambridge University Press.
McDonald, J. (2009). Using least squares and tobit in second stage dea efficiency
analyses. European Jounral of Operational Research 197, 792 – 798.
Mitchell, O. S. and E. S. Andrews (1981). Scale economies in private multi-employer
pension systems. Industrial and Labor Relations Review 34, 522–530.
OECD (2007). Pension at a glance 2007: Public policies across oecd countries.
Paris .
Pagan, A. and F. Vella (1989). Diagnostic tests for models based on individual data:
A survey. Journal of Applied Econometrics 4, 29 – 59.
Papke, L, E. and M. Wooldridge, J (1996). Econommetric methods for fractional
response variables with an application to 401(k) plan participation rates. Journal
of Applied Econometrics 11 (6), 619 – 632.
Ramalho, E, A., J. S. Ramalho, J, and D. Henriques, P (2010). Fractional
regression models for second stage dea efficiency analyses. Journal of Productivity
Analysis 34, 239 – 255.
Sathye, M. (2011). The impact of financial crisis on the efficiency of superannuation
funds: Evidence from australia. Journal of Law and Financial Management 10,
16–27.
Scholz, H. and M. Wilkens (2008). Interpreting sharpe ratios - the market climate
bias. Finance Letters, also available at: http://papers. ssrn. com/sol3/papers.cfm.
Simar, L. and P. Wilson (1999). Estimating and bootstrapping malmquist indices.
European Jounral of Operational Research 115, 459 – 471.
119
Simar, L. and P. Wilson (2000). Statistical inference in nonparametric fonriter
models: The state of the art. Jounral of Productivity Analysis 13, 49–78.
Sy, W. and K. Liu (2009). Investment performance ranking of superannuation funds.
Autralian Prudential Regulation Authority .
Thanassoulis, E. (1997). Duality in data envelopment analysis under constant
returns to scale. IMA journal of Mathematics Applied in Business and Industry 8,
253 – 266.
Tisonas, E. and T. Papadogonas (2006). Firm exit and technical inefficiency.
Empirical Economics 31(2), 535–548.
Vidler, S. (2004). Superannuation: Choice, competition and administration costs.
Journal of Australian Political Economy 53, 27 – 42.
Wheelock, D, C. and W. Wilson, P (2000). Why do banks disappear? the
determinants of u.s. bank failures and acquisitions. The Review of Economics
and Statistics 81(1), 127–138.
Wheelock, D. and P. Wilson (1999). Technical progress, inefficiency, and
productivity change in u.s. banking, 1984 - 1993. Journal of Money, Credit,
and Banking 31(2), 212 – 234.
120
Appendix
Average Efficiency by Fund Type
Table 8.1: Average Efficiency by Fund Type
year Corporate Industry Public Sector Retail
2004 0.80 0.74 0.79 0.712005 0.63 0.69 0.74 0.692006 0.64 0.65 0.78 0.652007 0.67 0.75 0.78 0.762008 0.19 0.03 0.07 0.142009 0.27 0.03 0.01 0.112010 0.65 0.59 0.68 0.662011 0.60 0.76 0.85 0.722012 0.32 0.30 0.51 0.29
TOBIT Proof
E(y|x) =0 · P(∑
βkxk + µ < 0)
+ E(y|x, 0 < y < 1) · P(
0 <∑
βkxk + µ < 1)
+ 1 · P(
1 <∑
βkxk + µ)
E(y|x) =0 +
∑ βkxk + σφ(−
∑βkxkσ
)− φ
(1−
∑βkxkσ
)Φ(
1−∑βkxkσ
)− Φ
(−
∑βkxkσ
)
· P(−∑
βkxk < µ < 1−∑
βkxk
)+ 1 · P
(1−
∑βkxk < µ
)
121
VaR Benchmarks
Table 8.2: Return Volatility Benchmark Index
Abreviation Asset Class Benchmark Index Source
AS Australian Listed Equities S&P/ASX 200 Merged Accumulation Index Bloomberg
OSH International Listed Equities MSCI TR NetWorld Ex-Australia Local Bloomberg
P Australian Listed Property S&P/ASX 200 Propoerty Merged Accumulation Index Bloomberg
PD Australian Direct Property Australian Mercer Unlisted Property Funds Index Pre-Tax Mercer Investment Consulting
AFI Australian Fixed Interst UBS Composite Bond Index All Maturities Bloomberg
OFI International Fixed Interest JP Morgan World ex-Aust $A Bloomberg
C Cash UBS Bank Bill Index Bloomberg
O Other UBS Bank Bill Index Bloomberg
VaR Benchmark Volatilities
Table 8.3: Benchmark Return Volatilities (Std.Dev)
Equity(AU) Equity(Int) Bonds(AU) Bonds(Int) Property Cash
2003 0.00457 0.00765 0.00984 0.00581 0.00677 0.00168
2004 0.00425 0.00902 0.00674 0.00576 0.00767 0.00011
2005 0.00619 0.00671 0.00604 0.00414 0.01003 0.00024
2006 0.00843 0.00915 0.00485 0.00428 0.00978 0.00031
2007 0.01073 0.01251 0.00509 0.00860 0.01528 0.00030
2008 0.02152 0.02823 0.01041 0.01825 0.03003 0.00079
2009 0.01325 0.01878 0.00835 0.01082 0.02504 0.00042
2010 0.00991 0.01473 0.00904 0.00872 0.01158 0.00034
2011 0.01238 0.01801 0.00700 0.00863 0.02115 0.00017
2012 0.00736 0.01153 0.00944 0.00548 0.01114 0.00048
2013 0.00850 0.00892 0.00852 0.00520 0.01070 0.00016
122
Conventional Linear Model Results
Table 8.4: OLS & TOBIT estimates
(1) (2)LPM TOBIT
Corporate Fund Dummy −0.0179 −0.0211∗
(0.0111) (0.0121)Industry Fund Dummy −0.0470∗∗∗ −0.0565∗∗∗
(0.0172) (0.0180)Public Sector Fund Dummy −0.0607∗∗∗ −0.0641∗∗∗
(0.0202) (0.0227)Public Offer Dummy −0.0308∗∗∗ −0.0293∗∗∗
(0.00781) (0.00820)Defined Benefit Dummy −0.00396 0.00640
(0.0149) (0.0181)Corporate × Share −0.120∗∗∗ −0.119∗∗∗
(0.0395) (0.0402)Public Sector × Share 0.0782∗∗∗ 0.0899∗∗∗
(0.0199) (0.0228)Industry × Share 0.0325 0.0349∗
(0.0199) (0.0207)Corporate × Choice 0.0198∗∗∗ 0.0199∗∗∗
(0.00367) (0.00404)Public Sector × Choice −0.0341∗∗∗ −0.0379∗∗∗
(0.0101) (0.0117)Industry × Choice −0.0184∗ −0.0177∗
(0.00972) (0.0100)Corporate × Exit −0.0323∗∗ −0.0354∗∗
(0.0136) (0.0151)Exiting Market Dummy 0.0278∗∗ 0.0326∗∗
(0.0126) (0.0139)Market Share −0.0830∗∗∗ −0.0817∗∗∗
(0.0129) (0.0132)Members in Default Option 0.0000870 0.000107
(0.0000782) (0.0000849)Herfindahl Index −0.0321∗∗∗ −0.0315∗∗∗
(0.00694) (0.00735)Investment Choices (log) −0.0180∗∗∗ −0.0201∗∗∗
Constant 1.053∗∗∗ 1.062∗∗∗
(0.0162) (0.0172)
σ 0.124∗∗∗
(0.00320)Year Fixed Effect Y es Y es
Adjusted R2 0.472Standard errors in parentheses∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
123
Fractional Response Model Results
Table 8.5: Generalised Linear Model Results
(1) (2) (3) (4)
Efficiency Score (VRS) Efficiency Score (VRS) Efficiency Score (VRS) Efficiency Score (VRS)
Corporate Fund Dummy 0.438∗∗∗ 0.165∗∗ 0.459∗∗∗ 0.0824∗
(0.145) (0.0696) (0.137) (0.0487)
Industry Fund Dummy −0.596∗∗∗ −0.288∗∗∗ −0.569∗∗∗ −0.203∗∗∗
(0.177) (0.0911) (0.165) (0.0668)
Public Sector Fund Dummy −0.674∗∗∗ −0.330∗∗∗ −0.629∗∗∗ −0.239∗∗∗
(0.240) (0.123) (0.214) (0.0926)
Public Offer Dummy −0.314∗∗∗ −0.151∗∗∗ −0.286∗∗∗ −0.104∗∗∗
(0.0764) (0.0388) (0.0714) (0.0286)
Defined Benefit Dummy −0.0582 −0.0176 −0.0763 −0.00630
(0.192) (0.0961) (0.181) (0.0684)
Corporate × Share −1.477∗∗∗ −0.796∗∗∗ −1.335∗∗∗ −0.634∗∗∗
(0.218) (0.131) (0.139) (0.147)
Public Sector × Share 0.423∗∗∗ 0.258∗∗∗ 0.297∗∗∗ 0.291∗∗∗
(0.119) (0.0699) (0.0890) (0.0726)
Industry × Share 0.178∗ 0.112∗ 0.105∗ 0.149∗∗
(0.0968) (0.0576) (0.0621) (0.0676)
Corporate × Choice 0.0910∗ 0.0667∗∗∗ 0.0718 0.0590∗∗∗
(0.0548) (0.0256) (0.0514) (0.0175)
Public Sector × Choice −0.243∗∗ −0.144∗∗ −0.194∗∗ −0.131∗∗∗
(0.108) (0.0567) (0.0916) (0.0444)
Industry × Choice −0.122 −0.0777∗ −0.0928 −0.0756∗∗
(0.0783) (0.0426) (0.0698) (0.0340)
Corporate × Exit −0.321 −0.177 −0.297 −0.127
(0.297) (0.136) (0.282) (0.0899)
Exiting Market Dummy 0.643∗∗ 0.272∗∗ 0.647∗∗∗ 0.163∗∗
(0.257) (0.119) (0.243) (0.0795)
Market Share −0.467∗∗∗ −0.276∗∗∗ −0.351∗∗∗ −0.297∗∗∗
(0.0676) (0.0383) (0.0333) (0.0488)
Members in Default Option 0.00202∗∗ 0.00107∗∗ 0.00173∗ 0.000841∗∗
(0.000968) (0.000484) (0.000902) (0.000348)
Herfindahl Index −0.365∗∗∗ −0.185∗∗∗ −0.345∗∗∗ −0.133∗∗∗
(0.0670) (0.0342) (0.0622) (0.0255)
Investment Choices (log) −0.182∗∗∗ −0.0952∗∗∗ −0.173∗∗∗ −0.0690∗∗∗
(0.0272) (0.0139) (0.0238) (0.0104)
Constant 3.891∗∗∗ 2.139∗∗∗ 3.825∗∗∗ 1.472∗∗∗
(0.184) (0.0897) (0.173) (0.0634)
Fixed Year Effect Y es Y es Y es Y es
Standard errors in parentheses
∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01
124
Recommended